US 8,805,684 B1Grant
Distributed speaker adaptation
Issue Date:2014-08-12
•22 Claims
•10 Drawing Sheets
Abstract
Automatic speech recognition (ASR) may be performed on received utterances. The ASR may be performed by an ASR module of a computing device (e.g., a client device). The ASR may include: generating feature vectors based on the utterances, updating the feature vectors based on feature-space speaker adaptation parameters, transcribing the utterances to text strings, and updating the feature-space speaker adaptation parameters based on the feature vectors. The transcriptions may be based, at least in part, on an acoustic model and the updated feature vectors. Updated speaker adaptation parameters may be received from another computing device and incorporated into the ASR module.
Metadata
Assignee
- Google Inc.
Inventors
- Petar Aleksic
- Xin Lei
Application Information
Application Number:US 13/653,804
Filing Date:2012-10-17
Priority Date:2012-05-31
Art Unit:2659
Classifications
IPC:
G10L15/00G10L15/06G10L25/00G10L21/00
Patent Drawings (10 sheets)
Description
CROSS-REFERENCE TO RELATED APPLICATION
[0001] This application is entitled to the benefit of provisional U.S. patent application Ser. No. 61/653,691, which is hereby incorporated by reference in its entirety.
BACKGROUND
[0002] Automatic speech recognition (ASR) technology can be used to map audio utterances to textual representations of those utterances. In some systems, ASR involves comparing characteristics of the audio utterances to an acoustic model of human voice. However, different speakers may exhibit different speech characteristics (e.g., pitch, accent, tempo, etc.). Consequently, the acoustic model may not perform well for all speakers.
SUMMARY
[0003] In a first example embodiment, automatic speech recognition (ASR) may be performed by an ASR module. The ASR may be performed on received utterances, where the ASR possibly includes generating feature vectors based on the utterances, updating the feature vectors based on feature-space speaker adaptation parameters, transcribing the utterances to text strings, and updating the feature-space speaker adaptation parameters based on the feature vectors. The transcriptions may be based, at least in part, on an acoustic model and the updated feature vectors. Updated speaker adaptation parameters may be received and incorporated into the ASR module.
[0004] A second example embodiment may include a non-transitory computer-readable storage medium, having stored thereon program instructions that, upon execution by a computing device, cause the computing device to perform operations in accordance with the first example embodiment.
[0005] A third example embodiment may include a computing device, comprising at least an ASR module and a communication module that operate in accordance with the first example embodiment.
[0006] A fourth example embodiment may include a user device and a computing device. The user device may contain an ASR module, wherein the ASR involves generating feature vectors based on the utterances, updating the feature vectors based on feature-space speaker adaptation parameters, transcribing the utterances to text strings, and updating the feature-space speaker adaptation parameters based on the feature vectors. The transcriptions may be based, at least in part, on an acoustic model and the updated feature vectors. The computing device may receive representations of one or more of the utterances transcribed by the user device and transmit updated speaker adaptation parameters, based on the representations, to the user device. The user device may incorporate the updated speaker adaptation parameters into the ASR module.
[0007] These as well as other aspects, advantages, and alternatives will become apparent to those of ordinary skill in the art by reading the following detailed description with reference where appropriate to the accompanying drawings. Further, it should be understood that the description provided in this summary section and elsewhere in this document is intended to illustrate the claimed subject matter by way of example and not by way of limitation.
BRIEF DESCRIPTION OF THE FIGURES
[0008] FIG. 1 depicts a distributed computing architecture, in accordance with an example embodiment.
[0009] FIG. 2A is a block diagram of a server device, in accordance with an example embodiment.
[0010] FIG. 2B depicts a cloud-based server system, in accordance with an example embodiment.
[0011] FIG. 3 depicts a block diagram of a client device, in accordance with an example embodiment.
[0012] FIG. 4 depicts an ASR system, in accordance with an example embodiment.
[0013] FIG. 5 depicts aspects of an acoustic model, in accordance with an example embodiment.
[0014] FIG. 6 depicts an ASR system search graph, in accordance with an example embodiment.
[0015] FIG. 7 depicts a form of distributed speaker adaptation, in accordance with an example embodiment.
[0016] FIG. 8 depicts another form of distributed speaker adaptation, in accordance with an example embodiment.
[0017] FIG. 9 is a flow chart, in accordance with an example embodiment.
DETAILED DESCRIPTION
1. Overview
[0018] ASR systems may include an acoustic model. The acoustic model may be speaker independent, in that it can represent the temporal and/or spectral characteristics of various sub-word sounds (e.g., phonemes) of a hypothetical “average” speaker. However, utterances from different speakers can have different qualities (e.g., different frequencies, tempos, accents, etc.). Thus, an acoustic model may perform reasonably well for most speakers, but may perform poorly when processing the utterances of some speakers. Particularly, speakers whose voices exhibit temporal and/or spectral characteristics that have not been appropriately represented in the acoustic model.
[0019] Speaker-dependent acoustic models can also be developed. These acoustic models are tuned to the speech characteristics of a particular speaker. However, developing a speaker-dependent acoustic model may involve collecting a great deal of accurately-transcribed utterances of this particular speaker. Thus, despite the error rates of speaker-dependent acoustic models being two to three times lower than those of speaker-independent acoustic models, speaker-dependent acoustic models may not always be possible or practical to develop.
[0020] Speaker adaptation can also be used with acoustic models. At a high level, speaker adaptation may involve the ASR system determining (i) the temporal and/or spectral characteristics of a particular speaker's voice, (ii) the difference between these characteristics and associated characteristics of the acoustic model, and (iii) developing a transform that maps the temporal and/or spectral characteristics of the particular speaker's voice to a representation that is closer to that of the acoustic model. This transform may then be applied to subsequent utterances received from the speaker, and the acoustic model may be applied to the result. Developing the transform and applying it to map the speaker's vocal characteristics to an acoustic model can be referred to as feature-space speaker adaptation.
[0021] Additionally or alternatively, speaker adaptation may involve adjusting the acoustic model itself based on the characteristics of the particular speaker's voice. For instance, once the ASR system has processed a sufficiently large set of samples of the speaker's utterances using an initial acoustic model, a new acoustic model can be derived from these utterances. The new acoustic model may be used in place of the initial acoustic model. This approach can be referred to as model-space speaker adaptation. Typically, more samples of the speaker's utterances and more computing resources are used to perform model-space speaker adaptation than are used to perform feature-space speaker adaptation.
[0022] ASR systems can be deployed in various environments. Some ASR systems are employed in a user device (e.g., a personal computer, tablet, or wireless communication device). A user speaks utterances into the device, and the ASR system in the device transcribes the utterances into one or more text strings. Other ASR systems are server-based. A user speaks an utterance into a client device, and the client device transmits the utterance (e.g., in an encoded form) to a server device. Then, the server device performs ASR on the utterance and transmits one or more text string mappings to the client device.
[0023] As the computing resources available to client devices continue to increase, aspects of ASR systems may be distributed in various ways between client and server devices. Since most users of ASR systems expect ASR to occur in real time or near real time, this distribution may occur in ways that leverage the computing resources available to each type of device.
[0024] For example, a disadvantage of using server-based ASR is that it uses communication between the client component and the server component. Thus, for example, when a cell phone operating as an ASR client function loses network connectivity, the ASR function may be unavailable. Client-based ASR can be usable even when the client device is unable to communicate with other devices. However, client devices (e.g., user devices) often have limited computing resources, which can affect the accuracy of the ASR. On the other hand, server-based ASR can leverage the computing resources available to server devices to perform ASR with higher accuracy (e.g., a lower word-error rate).
[0025] The embodiments described herein may be distributed in various ways between client and server devices. These embodiments can be deployed using client-based ASR systems or server-based ASR systems. Alternatively, aspects of these embodiments can be distributed in various ways between client and server devices to leverage the advantages of each.
[0026] For example, some variations of feature-space speaker adaptation can be used without reliance on extensive computing resources. Thus, in some embodiments, feature-space speaker adaptation may take place on client devices. Conversely, other variations of feature-space speaker adaptation may use more computing resources and therefore may take place on server devices. Similarly, variations of model-space speaker adaptation may use extensive computing resources and therefore may also take place on server devices.
[0027] Further, some possible embodiments involve a client device performing feature-space speaker adaptation locally to derive a set of feature-space speaker adaptation parameters. But, from time to time, the client device may receive new feature-space speaker adaptation parameters from a server device. Thus, the client device may use its computing resources to perform feature-space speaker adaptation, while the server device uses its potentially more extensive computing resources to perform feature-space speaker adaptation with greater accuracy. Therefore, in order to potentially improve the quality of speaker adaptation, the client device may replace its locally-derived feature-space speaker adaptation parameters with those from the server.
[0028] Likewise, from time to time, the client device may receive new model-space speaker adaptation parameters (e.g., a new acoustic model) from a server device. The new acoustic model may be adapted to the speaker using the client device. The client device may replace its acoustic model with the new, speaker-dependent acoustic model. Then, the client device may use the new acoustic model for ASR, but may continue performing local feature-space speaker adaptation.
[0029] The above processes, and example embodiments thereof, will be described in detail in Sections 5 and 6. However, in order to further embody ASR system implementations, the next three sections describe, respectively, example computing systems and devices that may support ASR systems, an overview of ASR system components and functions, and an overview of ASR system operation.
2. Example Communication System and Device Architecture for Supporting Automatic Speech Recognition
[0030] The methods, devices, and systems described herein can be implemented using client devices and/or so-called “cloud-based” server devices. Under various aspects of this paradigm, client devices, such as mobile phones, tablet computers, and/or desktop computers, may offload some processing and storage responsibilities to remote server devices. At least some of the time, these client services are able to communicate, via a network such as the Internet, with the server devices. As a result, applications that operate on the client devices may also have a persistent, server-based component. Nonetheless, it should be noted that at least some of the methods, processes, and techniques disclosed herein may be able to operate entirely on a client device or a server device.
[0031] Furthermore, the “server devices” described herein may not necessarily be associated with a client/server architecture, and therefore may also be referred to as “computing devices.” Similarly, the “client devices” described herein also may not necessarily be associated with a client/server architecture, and therefore may be interchangeably referred to as “user devices.” In some contexts, “client devices” may also be referred to as “computing devices.”
[0032] This section describes general system and device architectures for such client devices and server devices. However, the methods, devices, and systems presented in the subsequent sections may operate under different paradigms as well. Thus, the embodiments of this section are merely examples of how these methods, devices, and systems can be enabled.
[0033] A. Communication System
[0034] FIG. 1 is a simplified block diagram of a communication system 100, in which various embodiments described herein can be employed. Communication system 100 includes client devices 102, 104, and 106, which represent a desktop personal computer (PC), a tablet computer, and a mobile phone, respectively. Each of these client devices may be able to communicate with other devices via a network 108 through the use of wireline connections (designated by solid lines) and/or wireless connections (designated by dashed lines).
[0035] Network 108 may be, for example, the Internet, or some other form of public or private Internet Protocol (IP) network. Thus, client devices 102, 104, and 106 may communicate using packet-switching technologies. Nonetheless, network 108 may also incorporate at least some circuit-switching technologies, and client devices 102, 104, and 106 may communicate via circuit switching alternatively or in addition to packet switching.
[0036] A server device 110 may also communicate via network 108. Particularly, server device 110 may communicate with client devices 102, 104, and 106 according to one or more network protocols and/or application-level protocols to facilitate the use of network-based or cloud-based computing on these client devices. Server device 110 may include integrated data storage (e.g., memory, disk drives, etc.) and may also be able to access a separate server data storage 112. Communication between server device 110 and server data storage 112 may be direct, via network 108, or both direct and via network 108 as illustrated in FIG. 1 . Server data storage 112 may store application data that is used to facilitate the operations of applications performed by client devices 102, 104, and 106 and server device 110.
[0037] Although only three client devices, one server device, and one server data storage are shown in FIG. 1 , communication system 100 may include any number of each of these components. For instance, communication system 100 may comprise millions of client devices, thousands of server devices and/or thousands of server data storages. Furthermore, client devices may take on forms other than those in FIG. 1 .
[0038] B. Server Device
[0039] FIG. 2A is a block diagram of a server device in accordance with an example embodiment. In particular, server device 200 shown in FIG. 2A can be configured to perform one or more functions of server device 110 and/or server data storage 112. Server device 200 may include a user interface 202, a communication interface 204, processor 206, and data storage 208, all of which may be linked together via a system bus, network, or other connection mechanism 214.
[0040] User interface 202 may comprise user input devices such as a keyboard, a keypad, a touch screen, a computer mouse, a track ball, a joystick, and/or other similar devices, now known or later developed. User interface 202 may also comprise user display devices, such as one or more cathode ray tubes (CRT), liquid crystal displays (LCD), light emitting diodes (LEDs), displays using digital light processing (DLP) technology, printers, light bulbs, and/or other similar devices, now known or later developed. Additionally, user interface 202 may be configured to generate audible output(s), via a speaker, speaker jack, audio output port, audio output device, earphones, and/or other similar devices, now known or later developed. In some embodiments, user interface 202 may include software, circuitry, or another form of logic that can transmit data to and/or receive data from external user input/output devices.
[0041] Communication interface 204 may include one or more wireless interfaces and/or wireline interfaces that are configurable to communicate via a network, such as network 108 shown in FIG. 1 . The wireless interfaces, if present, may include one or more wireless transceivers, such as a BLUETOOTH® transceiver, a Wifi transceiver perhaps operating in accordance with an IEEE 802.11 standard (e.g., 802.11b, 802.11g, 802.11n), a WiMAX transceiver perhaps operating in accordance with an IEEE 802.16 standard, a Long-Term Evolution (LTE) transceiver perhaps operating in accordance with a 3rd Generation Partnership Project (3GPP) standard, and/or other types of wireless transceivers configurable to communicate via local-area or wide-area wireless networks. The wireline interfaces, if present, may include one or more wireline transceivers, such as an Ethernet transceiver, a Universal Serial Bus (USB) transceiver, or similar transceiver configurable to communicate via a twisted pair wire, a coaxial cable, a fiber-optic link or other physical connection to a wireline device or network.
[0042] Processor 206 may include one or more general purpose processors (e.g., microprocessors) and/or one or more special purpose processors (e.g., digital signal processors (DSPs), graphical processing units (GPUs), floating point processing units (FPUs), network processors, or application specific integrated circuits (ASICs)). Processor 206 may be configured to execute computer-readable program instructions 210 that are contained in data storage 208, and/or other instructions, to carry out various functions described herein.
[0043] Thus, data storage 208 may include one or more non-transitory computer-readable storage media that can be read or accessed by processor 206. The one or more computer-readable storage media may include volatile and/or non-volatile storage components, such as optical, magnetic, organic or other memory or disc storage, which can be integrated in whole or in part with processor 206. In some embodiments, data storage 208 may be implemented using a single physical device (e.g., one optical, magnetic, organic or other memory or disc storage unit), while in other embodiments, data storage 208 may be implemented using two or more physical devices.
[0044] Data storage 208 may also include program data 212 that can be used by processor 206 to carry out functions described herein. In some embodiments, data storage 208 may include, or have access to, additional data storage components or devices (e.g., cluster data storages described below).
[0045] C. Server Clusters
[0046] Server device 110 and server data storage device 112 may store applications and application data at one or more places accessible via network 108. These places may be data centers containing numerous servers and storage devices. The exact physical location, connectivity, and configuration of server device 110 and server data storage device 112 may be unknown and/or unimportant to client devices. Accordingly, server device 110 and server data storage device 112 may be referred to as “cloud-based” devices that are housed at various remote locations. One possible advantage of such “cloud-based” computing is to offload processing and data storage from client devices, thereby simplifying the design and requirements of these client devices.
[0047] In some embodiments, server device 110 and server data storage device 112 may be a single computing device residing in a single data center. In other embodiments, server device 110 and server data storage device 112 may include multiple computing devices in a data center, or even multiple computing devices in multiple data centers, where the data centers are located in diverse geographic locations. For example, FIG. 1 depicts each of server device 110 and server data storage device 112 potentially residing in a different physical location.
[0048] FIG. 2B depicts a cloud-based server cluster in accordance with an example embodiment. In FIG. 2B , functions of server device 110 and server data storage device 112 may be distributed among three server clusters 220A, 220B, and 220C. Server cluster 220A may include one or more server devices 200A, cluster data storage 222A, and cluster routers 224A connected by a local cluster network 226A. Similarly, server cluster 220B may include one or more server devices 200B, cluster data storage 222B, and cluster routers 224B connected by a local cluster network 226B. Likewise, server cluster 220C may include one or more server devices 200C, cluster data storage 222C, and cluster routers 224C connected by a local cluster network 226C. Server clusters 220A, 220B, and 220C may communicate with network 108 via communication links 228A, 228B, and 228C, respectively.
[0049] In some embodiments, each of the server clusters 220A, 220B, and 220C may have an equal number of server devices, an equal number of cluster data storages, and an equal number of cluster routers. In other embodiments, however, some or all of the server clusters 220A, 220B, and 220C may have different numbers of server devices, different numbers of cluster data storages, and/or different numbers of cluster routers. The number of server devices, cluster data storages, and cluster routers in each server cluster may depend on the computing task(s) and/or applications assigned to each server cluster.
[0050] In the server cluster 220A, for example, server devices 200A can be configured to perform various computing tasks of server device 110. In one embodiment, these computing tasks can be distributed among one or more of server devices 200A. Server devices 200B and 200C in server clusters 220B and 220C may be configured the same or similarly to server devices 200A in server cluster 220A. On the other hand, in some embodiments, server devices 200A, 200B, and 200C each may be configured to perform different functions. For example, server devices 200A may be configured to perform one or more functions of server device 110, and server devices 200B and server device 200C may be configured to perform functions of one or more other server devices. Similarly, the functions of server data storage device 112 can be dedicated to a single server cluster, or spread across multiple server clusters.
[0051] Cluster data storages 222A, 222B, and 222C of the server clusters 220A, 220B, and 220C, respectively, may be data storage arrays that include disk array controllers configured to manage read and write access to groups of hard disk drives. The disk array controllers, alone or in conjunction with their respective server devices, may also be configured to manage backup or redundant copies of the data stored in cluster data storages to protect against disk drive failures or other types of failures that prevent one or more server devices from accessing one or more cluster data storages.
[0052] Similar to the manner in which the functions of server device 110 and server data storage device 112 can be distributed across server clusters 220A, 220B, and 220C, various active portions and/or backup/redundant portions of these components can be distributed across cluster data storages 222A, 222B, and 222C. For example, some cluster data storages 222A, 222B, and 222C may be configured to store backup versions of data stored in other cluster data storages 222A, 222B, and 222C.
[0053] Cluster routers 224A, 224B, and 224C in server clusters 220A, 220B, and 220C, respectively, may include networking equipment configured to provide internal and external communications for the server clusters. For example, cluster routers 224A in server cluster 220A may include one or more packet-switching and/or routing devices configured to provide (i) network communications between server devices 200A and cluster data storage 222A via cluster network 226A, and/or (ii) network communications between the server cluster 220A and other devices via communication link 228A to network 108. Cluster routers 224B and 224C may include network equipment similar to cluster routers 224A, and cluster routers 224B and 224C may perform networking functions for server clusters 220B and 220C that cluster routers 224A perform for server cluster 220A.
[0054] Additionally, the configuration of cluster routers 224A, 224B, and 224C can be based at least in part on the data communication requirements of the server devices and cluster storage arrays, the data communications capabilities of the network equipment in the cluster routers 224A, 224B, and 224C, the latency and throughput of the local cluster networks 226A, 226B, 226C, the latency, throughput, and cost of the wide area network connections 228A, 228B, and 228C, and/or other factors that may contribute to the cost, speed, fault-tolerance, resiliency, efficiency and/or other design goals of the system architecture.
[0055] D. Client Device
[0056] FIG. 3 is a simplified block diagram showing some of the components of an example client device 300. By way of example and without limitation, client device 300 may be a “plain old telephone system” (POTS) telephone, a cellular mobile telephone, a still camera, a video camera, a fax machine, an answering machine, a computer (such as a desktop, notebook, or tablet computer), a personal digital assistant (PDA), a home automation component, a digital video recorder (DVR), a digital TV, a remote control, or some other type of device equipped with one or more wireless or wired communication interfaces.
[0057] As shown in FIG. 3 , client device 300 may include a communication interface 302, a user interface 304, a processor 306, and data storage 308, all of which may be communicatively linked together by a system bus, network, or other connection mechanism 310.
[0058] Communication interface 302 functions to allow client device 300 to communicate, using analog or digital modulation, with other devices, access networks, and/or transport networks. Thus, communication interface 302 may facilitate circuit-switched and/or packet-switched communication, such as POTS communication and/or IP or other packetized communication. For instance, communication interface 302 may include a chipset and antenna arranged for wireless communication with a radio access network or an access point. Also, communication interface 302 may take the form of a wireline interface, such as an Ethernet, Token Ring, or USB port. Communication interface 302 may also take the form of a wireless interface, such as a Wifi, BLUETOOTH®, global positioning system (GPS), or wide-area wireless interface (e.g., WiMAX or LTE). However, other forms of physical layer interfaces and other types of standard or proprietary communication protocols may be used over communication interface 302. Furthermore, communication interface 302 may comprise multiple physical communication interfaces (e.g., a Wifi interface, a BLUETOOTH®) interface, and a wide-area wireless interface).
[0059] User interface 304 may function to allow client device 300 to interact with a human or non-human user, such as to receive input from a user and to provide output to the user. Thus, user interface 304 may include input components such as a keypad, keyboard, touch-sensitive or presence-sensitive panel, computer mouse, trackball, joystick, microphone, still camera and/or video camera. User interface 304 may also include one or more output components such as a display screen (which, for example, may be combined with a presence-sensitive panel), CRT, LCD, LED, a display using DLP technology, printer, light bulb, and/or other similar devices, now known or later developed. User interface 304 may also be configured to generate audible output(s), via a speaker, speaker jack, audio output port, audio output device, earphones, and/or other similar devices, now known or later developed. In some embodiments, user interface 304 may include software, circuitry, or another form of logic that can transmit data to and/or receive data from external user input/output devices. Additionally or alternatively, client device 300 may support remote access from another device, via communication interface 302 or via another physical interface (not shown).
[0060] Processor 306 may comprise one or more general purpose processors (e.g., microprocessors) and/or one or more special purpose processors (e.g., DSPs, GPUs, FPUs, network processors, or ASICs). Data storage 308 may include one or more volatile and/or non-volatile storage components, such as magnetic, optical, flash, or organic storage, and may be integrated in whole or in part with processor 306. Data storage 308 may include removable and/or non-removable components.
[0061] Processor 306 may be capable of executing program instructions 318 (e.g., compiled or non-compiled program logic and/or machine code) stored in data storage 308 to carry out the various functions described herein. Therefore, data storage 308 may include a non-transitory computer-readable storage medium, having stored thereon program instructions that, upon execution by client device 300, cause client device 300 to carry out any of the methods, processes, or functions disclosed in this specification and/or the accompanying drawings. The execution of program instructions 318 by processor 306 may result in processor 306 using data 312.
[0062] By way of example, program instructions 318 may include an operating system 322 (e.g., an operating system kernel, device driver(s), and/or other modules) and one or more application programs 320 (e.g., address book, email, web browsing, social networking, and/or gaming applications) installed on client device 300. Similarly, data 312 may include operating system data 316 and application data 314. Operating system data 316 may be accessible primarily to operating system 322, and application data 314 may be accessible primarily to one or more of application programs 320. Application data 314 may be arranged in a file system that is visible to or hidden from a user of client device 300.
[0063] Application programs 320 may communicate with operating system 322 through one or more application programming interfaces (APIs). These APIs may facilitate, for instance, application programs 320 reading and/or writing application data 314, transmitting or receiving information via communication interface 302, receiving or displaying information on user interface 304, and so on.
[0064] In some vernaculars, application programs 320 may be referred to as “apps” for short. Additionally, application programs 320 may be downloadable to client device 300 through one or more online application stores or application markets. However, application programs can also be installed on client device 300 in other ways, such as via a web browser or through a physical interface (e.g., a USB port) on client device 300.
3. Example Automatic Speech Recognition System Overview
[0065] Before describing speaker adaptation in detail, it may be beneficial to understand overall ASR system operation. Thus, this section describes ASR systems in general, including how the components of an ASR system may interact with one another in order to facilitate speech recognition, and how some of these components may be trained.
[0066] FIG. 4 depicts an example ASR system. At run-time, the input to the ASR system may include an utterance 400, and the output may include one or more text strings and possibly associated confidence levels 414. The components of the ASR system may include a feature analysis module 402 that produces feature vectors 404, a pattern classification module 406, an acoustic model 408, a dictionary 410, and a language model 412. Pattern classification module 406 may incorporate various aspects of acoustic model 408, dictionary 410, and language model 412.
[0067] It should be noted that the discussion in this section, and the accompanying figures, are presented for purposes of example. Other ASR system arrangements, including different components, different relationships between the components, and/or different processing, may be possible.
[0068] A. Feature Analysis Module
[0069] Feature analysis module 402 may receive utterance 400. This utterance may include an analog or digital representation of human speech, and may possibly contain background noise as well. Feature analysis module 402 may convert utterance 400 to a sequence of one or more feature vectors 404. Each of feature vectors 404 may include temporal and/or spectral representations of the acoustic features of at least a portion of utterance 400. For instance, a feature vector may include mel-frequency cepstrum coefficients of such a portion.
[0070] The mel-frequency cepstrum coefficients may represent the short-term power spectrum of a portion of utterance 400. They may be based on, for example, a linear cosine transform of a log power spectrum on a nonlinear mel scale of frequency. (A mel scale may be a scale of pitches subjectively perceived by listeners to be about equally distant from one another, even though the actual frequencies of these pitches are not equally distant from one another.)
[0071] To derive these coefficients, feature analysis module 402 may sample and quantize utterance 400, divide it into overlapping or non-overlapping frames of s milliseconds, and perform spectral analysis on the frames to derive the spectral components of each frame. Feature analysis module 402 may further perform noise removal and convert the standard spectral coefficients to mel-frequency cepstrum coefficients, and then calculate first-order and second-order cepstral derivatives of the mel-frequency cepstrum coefficients.
[0072] The first-order cepstral coefficient derivatives may be calculated based on the slopes of linear regressions performed over windows of two or more consecutive frames. The second-order cepstral coefficient derivatives may be calculated based on the slopes of linear regressions performed over windows of two or more consecutive sets of first-order cepstral coefficient derivatives. However, there may be other ways of calculating the first-order and second-order cepstral coefficient derivatives.
[0073] In some embodiments, one or more frames of utterance 400 may be represented by a feature vector of mel-frequency cepstrum coefficients, first-order cepstral coefficient derivatives, and second-order cepstral coefficient derivatives. For example, the feature vector may contain 13 coefficients, 13 first-order derivatives, and 13 second-order derivatives, therefore having a length of 39. However, feature vectors may use different combinations of features in other possible embodiments.
[0074] B. Pattern Classification Module
[0075] Pattern classification module 406 may receive a sequence of feature vectors 404 from feature analysis module 402 and produce, as output, one or more text string transcriptions 414 of utterance 400. Each transcription 414 may be accompanied by a respective confidence level indicating an estimated likelihood that the transcription is correct (e.g., 80% confidence, 90% confidence, etc.).
[0076] To produce this output, pattern classification module 406 may include, or incorporate aspects of acoustic model 408, dictionary 410, and/or language model 412. In some embodiments, pattern classification module 406 may also use a search graph that represents sequences of word or sub-word acoustic features that appear in spoken utterances. The behavior of pattern classification module 406 will be described below in the context of these modules.
[0077] C. Acoustic Model
[0078] Acoustic model 408 may determine probabilities that a particular sequence of feature vectors 404 were derived from particular sequence of spoken words and/or sub-word sounds. This may involve mapping sequences of feature vectors to one or more phonemes, and then mapping sequences of phonemes to one or more words.
[0079] A phoneme may be considered to be the smallest segment of an utterance that encompasses a meaningful contrast with other segments of utterances. Thus, a word typically includes one or more phonemes. For purposes of simplicity, phonemes may be thought of as utterances of letters, but this is not a perfect analogy, as some phonemes may present multiple letters. An example phonemic spelling for the American English pronunciation of the word “cat” is /k/ /ae/ /t/, consisting of the phonemes /k/, /ae/, and /t/. Another example phonemic spelling for the word “dog” is /d/ /aw/ /g/, consisting of the phonemes /d/, /aw/, and /g/.
[0080] Different phonemic alphabets exist, and these alphabets may have different textual representations for the various phonemes therein. For example, the letter “a” may be represented by the phoneme /ae/ for the sound in “cat,” by the phoneme /ey/ for the sound in “ate,” and by the phoneme /ah/ for the sound in “beta.” Other phonemic representations are possible.
[0081] Common phonemic alphabets for American English contain about 40 distinct phonemes. Each of these phonemes may be associated with a different distribution of feature vector values. Thus, acoustic model 408 may be able to estimate the phoneme(s) in a feature vector by comparing the feature vector to the distributions for each of the 40 phonemes, and finding one or more phonemes that are most likely represented by the feature vector.
[0082] One way of doing so is through use of a hidden Markov model (HMM). An HMM may model a system as a Markov process with unobserved (i.e., hidden) states. Each HMM state may be represented as a multivariate Gaussian distribution that characterizes the statistical behavior of the state. Additionally, each state may also be associated with one or more state transitions that specify the probability of making a transition from the current state to another state.
[0083] When applied to an ASR system, the combination of the multivariate Gaussian distribution and the state transitions for each state may define a time sequence of feature vectors over the duration of one or more phonemes. Alternatively or additionally, the HMM may model the sequences of phonemes that define words. Thus, some HMM-based acoustic models may also consider phoneme context when a mapping a sequence of feature vectors to one or more words.
[0084] FIG. 5 depicts an example HMM-based acoustic model 500. Acoustic model 500 defines a sequence of phonemes that make up the word “cat.” Thus, each phoneme is represented by a 3-state HMM with an initial state, a middle state, and an end state representing the statistical characteristics at the beginning of phoneme, the middle of the phoneme, and the end of the phoneme, respectively. Each state (e.g., state /k/1, state /k/2, etc.) may represent a phoneme and may include one or more transitions.
[0085] Acoustic model 500 may represent a word by concatenating the respective 3-state HMMs for each phoneme in the word together, with appropriate transitions. These concatenations may be performed based on information in dictionary 410, as discussed below. In some implementations, more or fewer states per phoneme may be used in an acoustic model.
[0086] An acoustic model may be trained using recordings of each phoneme in numerous contexts (e.g., various words and sentences) so that a representation for each of the phoneme's states can be obtained. These representations may encompass the multivariate Gaussian distributions discussed above.
[0087] In order to train the acoustic model, a possibly large number of utterances containing spoken phonemes may each be associated with transcriptions. These utterances may be words, sentences, and so on, and may be obtained from recordings of everyday speech or some other source. The transcriptions may be high-accuracy automatic or manual (human-made) text strings of the utterances.
[0088] The utterances may be segmented according to their respective transcriptions. For instance, training of the acoustic models may involve segmenting the spoken strings into units (e.g., using either a Baum-Welch and/or Viterbi alignment method), and then using the segmented utterances to build distributions for each phoneme state.
[0089] Consequently, as more data (utterances and their associated transcriptions) are used for training, a more accurate acoustic model is expected to be produced. However, even a well-trained acoustic model may have limited accuracy when used for ASR in a domain for which it was not trained. For instance, if an acoustic model is trained by utterances from a number of speakers of American English, this acoustic model may perform well when used for ASR of American English, but may be less accurate when used for ASR of, e.g., British English.
[0090] Also, if an acoustic model is trained using utterances from a number of speakers, it will likely end up representing each phoneme as a statistical average of the pronunciation of this phoneme across all of the speakers. Thus, an acoustic model trained in this fashion may represent the pronunciation and usage of a hypothetical average speaker, rather than any particular speaker.
[0091] For purposes of simplicity, throughout this specification and the accompanying drawings, it is assumed that acoustic models represent phonemes as context-dependent phonemic sounds. However, acoustic models that use other types of representations are within the scope of the embodiments herein.
[0092] D. Dictionary
[0093]
| TABLE 1 | ||
|---|---|---|
| Word | Phonemic Interpretation | |
| cat | /k/ /ac/ /t/ | |
| and | /ay/ /n/ /d/ | |
| dog | /d/ /aw/ /g/ | |
[0094] As noted above, dictionary 410 may define a pre-established mapping between phonemes and words. This mapping may include a list of tens or hundreds of thousands of phoneme-pattern-to-word mappings. Thus, in some embodiments, dictionary 410 may include a lookup table, such as Table 1. Table 1 illustrates how dictionary 410 may list the phonemic sequences that pattern classification module 406 uses for the words that the ASR system is attempting to recognize. Therefore, dictionary 410 may be used when developing the phonemic state representations of words that are illustrated by acoustic model 500.
[0095] E. Language Model
[0096] Language model 412 may assign probabilities to sequences of phonemes or words, based on the likelihood of that sequence of phonemes or words occurring in an input utterance to the ASR system. Thus, for example, language model 412 may define the conditional probability of wn (the nth word in a phrase transcribed from an utterance), given the values of the pattern of n−1 previous words in the phrase. More formally, language model 412 may define
P(wn/w1, w2, . . . , wn−1)
P(wn/w1, w2, . . . , wn−1)
[0097] In general, a language model may operate on n-grams, which, for example, may be sequences of n phonemes or words that are represented in pattern classification module 406. In practice, language models with values of n greater than 5 are rarely used because of their computational complexity, and also because smaller n-grams (e.g., 3-grams, which are also referred to as tri-grams) tend to yield acceptable results. In the example described below, tri-grams are used for purposes of illustration. Nonetheless, any value of n may be may be used with the embodiments herein.
[0098] Language models may be trained through analysis of a corpus of text strings. This corpus may contain a large number of words, e.g., hundreds, thousands, millions or more. These words may be derived from utterances spoken by users of an ASR system and/or from written documents. For instance, a language model can be based on the word patterns occurring in human speech, written text (e.g., emails, web pages, reports, academic papers, word processing documents, etc.), and so on.
[0099] From such a corpus, tri-gram probabilities can be estimated based on their respective number of appearances in the training corpus. In other words, if C(w1, w2, w3) is the number of occurrences of the word pattern w1, w2, w3 in the corpus, then
[0100]
[0101] Thus, a language model may be represented as a table of conditional probabilities. Table 2 illustrates a simple example of such a table that could form the basis of language model 406. Particularly, Table 2 contains tri-gram conditional probabilities.
[0102]
| TABLE 2 | |
|---|---|
| Tri-gram Conditional Probabilities | |
| P(dog|cat, and) = 0.50 | |
| P(mouse|cat, and) = 0.35 | |
| P(bird|cat, and) = 0.14 | |
| P(fiddle|cat, and) = 0.01 | |
[0103] For the 2-gram prefix “cat and,” Table 2 indicates that, based on the observed occurrences in the corpus, 50% of the time the next 1-gram is “dog.” Likewise, 35% of the time, the next 1-gram is “mouse,” 14% of the time the next 1-gram is “bird,” and 1% of the time the next 1-gram is “fiddle.” Clearly, in a fully-trained ASR system, the language model would contain many more entries, and these entries would include more than just one 2-gram prefix.
[0104] Nonetheless, using the observed frequencies of word patterns from a corpus of speech (and/or from other sources) is not perfect, as some acceptable tri-grams may not appear in the corpus, and may therefore be assigned a probability of zero. Consequently, when given a zero-probability tri-gram at run time, the language model may instead attempt to map this tri-gram to a different tri-gram associated with a non-zero probability.
[0105] In order to reduce this likelihood, the language model may be smoothed so that zero-probability tri-grams have small non-zero probabilities, and the probabilities of the tri-grams in the corpus are reduced accordingly. In this way, tri-grams not found in the corpus can still be recognized by the language model.
4. Example Automatic Speech Recognition System Operation
[0106] Once acoustic model 408 and language model 412 are appropriately trained, feature analysis model 402 and pattern classification module 406 may be used to perform ASR. Provided with an input utterance, the ASR system can search the space of valid word sequences from the language model to find the word sequence with the maximum likelihood of having been spoken in the utterance. A challenge with doing so is that the size of the search space can be quite large, and therefore performing this search may take an excessive amount computing resources (e.g., processing time and memory utilization). Nonetheless, there are some heuristic techniques that can be used to reduce the complexity of the search, potentially by orders of magnitude.
[0107] For instance, a finite state transducer (FST) can be used to compactly represent multiple phoneme patterns that map to a single word. Some words, such as “data,” “either,” “tomato,” and “potato,” have multiple pronunciations. The phoneme sequences for these pronunciations can be represented in a single FST per word.
[0108] This process of creating efficient phoneme-level FSTs can be carried out for each word in dictionary 410, and the resulting word FSTs can be combined into sentence FSTs using the language model 412. Ultimately, a very large network of states for phonemes, words, and sequences of words can be developed and represented in a compact search graph.
[0109] FIG. 6 contains an example search graph 600. In order to be illustrative, search graph 600 is much smaller and less complex than a search graph that would be used in an actual ASR system. Particularly, search graph 600 was trained with only the five input utterances, “catapult,” “cat and mouse,” “cat and dog,” “cat,” and “cap.”
[0110] Each circle in search graph 408 may represent a state associated with the processing of an input utterance that has been mapped to phonemes. For purposes of simplicity, each phoneme in search graph 600 is represented with a single state rather than multiple states. Also, self-transitions are omitted from search graph 600 in order to streamline FIG. 6 .
[0111] The states in search graph 600 are named based on the current phoneme context of the input utterance, using the format “x[y]z” to indicate that the current phoneme being considered, y, has a left-context of the phoneme x and a right context of the phoneme z. In other words, the state “x[y]z” indicates a point in processing an utterance in which the current phoneme being considered is y, the previously phoneme in the utterance is x, and the next phoneme in the utterance is z. The beginning of an utterance and the end of an utterance are represented by the “#” character, and also may be referred to as null phonemes.
[0112] Terminal states may be represented by a recognized word or phrase in quotes. Search graph 600 includes five terminal states, representing recognition of the words or phrases “catapult,” “cat and mouse,” “cat and dog,” “cat,” and “cap.”
[0113] Transitions from one state to another may represent an observed ordering of phonemes in the corpus. For instance, the state “#[k]ae” represents the recognition of a “k” phoneme with a left context of a null phoneme and a right context of an “ae” phoneme. There are two transitions from the state “#[k]ae”—one for which the next phoneme (the phoneme after the “ae”) is a “t” and another for which the next phoneme is a “p.”
[0114] Based on acoustic model 408, dictionary 410, and language model 412, costs may be assigned to one or more of the states and/or transitions. For example, if a particular phoneme pattern is rare, a transition to a state representing that phoneme pattern may have a higher cost than a transition to a state representing a more common phoneme pattern. Similarly, the conditional probabilities from the language model (see Table 2 for examples) may also be used to assign costs to states and/or transitions. For instance, in Table 2, given a phrase with the words “cat and,” the conditional probability of the next word in the phrase being “dog” is 0.5, while the conditional probability of the next word in the phrase being “mouse” is 0.35. Therefore, the transition from state “ae[n]d” to state “n[d]m” may have a higher cost than the transition from state “ae[n]d” to state “n[d]d.”
[0115] Search graph 600, including any states, transitions between states, and associated costs therein, may be used to estimate text string transcriptions for new input utterances. For example, pattern classification module 406 may determine a sequence of one or more words that match an input utterance based on search graph 600. Formally, pattern classification module 406 may attempt to find
w*=argmaxwP(a|w)P(w)
where a is a stream of feature vectors derived from the input utterance, P(a|w) represents the probability of those feature vectors being produced by a word sequence w, and P(w) is the probability assigned to w by language model 412. For example, P(w) may be based on n-gram conditional probabilities as discussed above, as well as other factors. The function argmaxw may return the value of w that maximizes P(a|w)P(w).
w*=argmaxwP(a|w)P(w)
where a is a stream of feature vectors derived from the input utterance, P(a|w) represents the probability of those feature vectors being produced by a word sequence w, and P(w) is the probability assigned to w by language model 412. For example, P(w) may be based on n-gram conditional probabilities as discussed above, as well as other factors. The function argmaxw may return the value of w that maximizes P(a|w)P(w).
[0116] To find text strings that may match utterance 400, pattern classification module 406 may attempt to find paths from an initial state in search graph 600 to a terminal state in search graph 600 based on feature vectors 404. This process may involve pattern classification module 406 performing a breadth-first search, depth-first search, beam search, or some other type of search on search graph 600. Pattern classification module 406 may assign a total cost to one or more paths through search graph 600 based on costs associated with the states and/or transitions of associated with each path. Some of these costs may reflect, for instance, a confidence level that a particular segment of the utterance maps to a particular sequence of phonemes in the path.
[0117] As an example, suppose that utterance 400 is the phrase “cat and dog.” In a possible scenario, pattern classification module 406 would step through search graph 600 phoneme by phoneme to find the path beginning with initial state “#[k]ae” and ending with terminal state “cat and dog.” Pattern classification module 406 may also find one or more additional paths through search graph 600. For example, pattern classification module 406 may also associate utterance 400 with the path with initial state “#[k]ae” and ending with terminal state “cat and mouse,” and with the path with initial state “#[k]ae” and ending with terminal state “catapult.” Nonetheless, pattern classification module 406 may assign a lower cost to the path with terminal state “cat and dog” than to other paths. Consequently, the path with terminal state “cat and dog” may be selected as the “best” transcription for the input utterance.
[0118] It should be understood that ASR systems can operated in many different ways. The embodiments described above are presented for purposes of illustration and may not be the only way in which an ASR system operates.
5. Examples of Distributed Speaker Adaptation Functions
[0119] As noted above, acoustic models are typically trained with utterances from multiple speakers in multiple environments. As a result, a given acoustic model may represent a hypothetical average speaker, and might not perform well when applied to utterances from a speaker whose vocal characteristics differ from those of the hypothetical average speaker. Therefore, ASR systems may attempt to compensate for these differences through speaker adaptation.
[0120] Speaker adaptation functions may be distributed in a variety of ways between client devices and server devices. FIG. 7 depicts an example embodiment in which feature-space speaker adaptation is performed in client devices and model-space speaker adaptation is performed in server devices.
[0121] The client device in FIG. 7 includes feature analysis module 402, feature-space speaker adaptation module 700, and pattern classification module 406. Although depicted as a separate component, feature-space speaker adaptation module 700 may be part of feature analysis module 402. Feature analysis model 402 and pattern classification module 406 may perform as described in the previous sections, but may also include any additional functions described below. While FIG. 7 depicts pattern classification model 406 as part of the client device, pattern classification module 406 may be part of the server device instead.
[0122] The server device may include model-space speaker adaptation module 702. The server device may also include other ASR system components, such as but not limited to pattern classification module 406. If the server device includes pattern classification module 406, the server device may include more extensive versions of the acoustic model, dictionary, and/or language model than the versions of these ASR components located on the client device.
[0123] Although not shown in FIG. 7 , the server device may also include profiles (e.g., information, settings, and/or configurations) for a number of individual speakers and/or client devices. Each profile may include speaker adaptation parameters associated with an individual speaker and/or client device.
[0124] Regardless, at step 704, feature analysis module 402 may receive an utterance and produce one or more feature vectors based on the utterance. As noted above, each of the feature vectors may be of a particular length (e.g., n entries), and may include representations of the temporal and/or spectral acoustic features of at least a portion of the utterance.
[0125] At step 706, feature-space speaker adaptation module 700 may the receive the feature vector(s). At step 708, feature-space speaker adaptation module 700 may update its feature-space speaker adaptation parameters based on the received feature vector(s).
[0126] In some embodiments, the feature-space speaker adaptation parameters may take the form of a matrix, for instance, an n×n matrix. In order to perform speaker adaptation, feature-space speaker adaptation module 700 may multiply each feature vector it receives by the matrix, resulting in updated feature vectors. These updated feature vectors may be transmitted to pattern classification module 406.
[0127] Thus, the acoustic model used by pattern classification module 406 may be speaker-independent, and feature-space speaker adaptation module 700 may use the matrix to adapt speaker-dependent feature vectors so that they are more likely to be properly recognized by the acoustic model. In some cases, the matrix may be a diagonal matrix (i.e., for each entry (i,j) in the matrix, the entry takes on a non-zero value if i is equal to j, but takes on a value of zero if i is not equal to j). Since at least half of the entries in a 2×2 or greater diagonal matrix contain zeroes, less computation is required to multiply a feature vector by a diagonal matrix than a non-diagonal matrix. (Herein, a non-diagonal matrix refers to a matrix in which at least one entry for which i is not equal to j contains a non-zero value.) Thus, the client device's possibly limited computational resources are conserved.
[0128] The feature-space speaker adaptation parameters may be updated in a number of ways, and though the use of a number of techniques. As one example, a feature-space maximum likelihood linear regression (fMLLR) technique may be used. Suppose that
Ô(t)=AO(t)+b
where O(t) is the feature vector at time t, Ô(t) is the adapted (e.g., transformed) feature vector at time t, A is a matrix and b is a bias factor. This equation can be re-written as
Ô(t)=WE(t)
where E(t) is the extended feature vector [1, O(t)]T, and W=[bT,AT]T. Thus, W is a matrix that may represent the feature-space speaker adaptation parameters, and therefore W may be applied to feature vectors received from feature analysis module 402.
Ô(t)=AO(t)+b
where O(t) is the feature vector at time t, Ô(t) is the adapted (e.g., transformed) feature vector at time t, A is a matrix and b is a bias factor. This equation can be re-written as
Ô(t)=WE(t)
where E(t) is the extended feature vector [1, O(t)]T, and W=[bT,AT]T. Thus, W is a matrix that may represent the feature-space speaker adaptation parameters, and therefore W may be applied to feature vectors received from feature analysis module 402.
[0129] In order to update W, any appropriate MLLR technique, or another type of machine learning technique, may be used. Particularly, W may be estimated iteratively based on the k previously-received feature vectors. Once W is determined, it may be used with subsequent feature vectors. In some embodiments, W may be updated once for each feature vector received. In other embodiments, W may be updated after a group of one or more feature vectors is received.
[0130] At step 710, the adapted feature vectors may be received at pattern classification module 406. At step 712, pattern classification module 406 may transcribe the adapted feature vectors to produce one or more possible text string transcriptions of the utterance. At least some of these text strings may be presented to a user of the client device as possible transcriptions of the utterance.
[0131] At step 714, feature analysis module 402 (or some other module of the client device), may transmit a set of recent utterances and/or feature vectors derived from these utterances to model-space speaker adaptation module 702. At step 716, model-space speaker adaptation module 702 may update the acoustic model that it uses for the client device.
[0132] In some embodiments, step 716 may involve the server device updating a speaker-dependent acoustic model that it has adapted for the client device. On the other hand, possibly if the server device has not yet adapted an acoustic model for the client device, step 716 may involve the server device deriving a new acoustic model for the client device. Regardless of how it is generated, the acoustic model for the client device may be a speaker-dependent acoustic model that is adapted to the characteristics of a speaker that has used the client device for ASR.
[0133] At step 718, the server device may transmit the new acoustic model to the client device. At step 720, the client device may incorporate the new acoustic model into pattern classification module 406. In some implementations, incorporating the new acoustic model may involve the client device replacing its existing acoustic model with the new acoustic model.
[0134] Alternatively or in addition to a server device deriving and transmitting updated model-space feature adaptation parameters to a client device, a server device may derive and transmit updated feature-space speaker adaptation parameters to a client device. Given that server devices typically have more processing power than client devices, feature-space speaker adaptation performed by a server device may result in better speaker adaptation parameters than when feature-space speaker adaptation is performed only on the client device.
[0135] FIG. 8 illustrates such an embodiment. Similar to the embodiment of FIG. 7 , the client device may include feature analysis module 402, feature-space speaker adaptation module 700, and pattern classification module 406. However, the server device includes feature-space speaker adaptation module 800. Additionally, as was the case for the embodiment illustrated by FIG. 7 , the server device may also include a profile for a number of individual speakers and/or client devices. This profile may include speaker adaptation parameters associated with each individual speaker and/or client device.
[0136] At step 802, feature analysis module 402 may receive an utterance and produce one or more feature vectors based on the utterance. At step 804, feature-space speaker adaptation module 700 may receive feature vector(s).
[0137] At step 806, feature-space speaker adaptation module 700 may update its feature-space speaker adaptation parameters based on the received feature vector(s). As was the case for the embodiment illustrated by FIG. 7 , the feature-space speaker adaptation parameters may include a diagonal or non-diagonal matrix that is used to adapt subsequent feature vector(s).
[0138] At step 808, feature-space speaker adaptation module may transmit the adapted feature vector(s) to pattern classification module 406. Based on one or more sets of feature vectors that were derived from the utterance, at step 810, pattern classification module 406 may transcribe these feature vectors to produce one or more possible text string transcriptions of the utterance. At least some of these text strings may be presented to a user of the client device as possible transcriptions of the utterance.
[0139] At step 812, feature analysis module 402 (or some other module of the client device), may transmit a set of recent utterances and/or feature vectors derived from these utterances to feature-space speaker adaptation module 800 on the server device. At step 814, feature-space speaker adaptation module 800 may update the server device's feature-space speaker adaptation parameters for the client device.
[0140] These updated feature-space speaker adaptation parameters may include a more extensive and/or complicated set of feature-space speaker adaptation parameters than those derived by the client device. For example, the updated feature-space speaker adaptation parameters derived by the server device may include an n×n non-diagonal matrix, whereas those derived by the client device may include an n×n diagonal matrix.
[0141] At step 816, the server device may transmit updated feature-space speaker adaptation parameters to the client device. At step 818, the client device may incorporate the updated feature-space speaker adaptation parameters into its feature-space speaker adaptation module. In some implementations, incorporating the updated feature-space speaker adaptation parameters may involve the client device replacing its existing feature-space speaker adaptation parameters with the updated feature-space speaker adaptation parameters.
[0142] FIG. 7 illustrates, among other features, a server device adapting an acoustic model on behalf of a client device and providing the updated (or new) acoustic model to the client device. FIG. 8 illustrates, among other features, a server device adapting feature-space speaker adaptation parameters on behalf of a client device and providing the updated (or new) feature-space speaker adaptation parameters to the client device. For both of these embodiments, after the updated acoustic model or the updated feature-space speaker adaptation parameters are incorporated into the client device, the client device may continue to perform local feature-space speaker adaptation. In this way, the server device may assist the client device by performing speaker adaptation periodically based on large amounts of data. On the other hand, the client device may perform speaker adaptation on a continuous or substantially-continuous fashion based on smaller amounts of data.
[0143] It should be understood that the embodiments illustrated in FIGS. 7 and 8 are for purposes of example. Other embodiments may be possible, including variations that change the order and/or content of the steps of FIGS. 7 and 8 . In some embodiments, more or fewer steps may be employed. Further, these embodiments may be combined, in part or in whole. For instance, a server device may adapt both acoustic models and feature-space speaker adaptation parameters on behalf of client devices, and transmit these updated model-space and feature-space speaker adaptation parameters to client devices.
6. Example Operations
[0144] FIG. 9 is a flow chart of an example embodiment. The steps illustrated by this flow chart may be carried out by various computing devices, such as client device 300, server device 200 and/or server cluster 220A. Aspects of some individual steps may be distributed between multiple computing devices.
[0145] At step 900, an ASR module of a user device (e.g., a client device) may perform ASR on received utterances. The ASR module may include one or more of a feature analysis module, a feature-space speaker adaptation module, and a pattern classification module. The ASR may include generating feature vectors based on the utterances, updating the feature vectors based on feature-space speaker adaptation parameters, transcribing the utterances to text strings, and updating the feature-space speaker adaptation parameters based on the feature vectors. The transcriptions may be based at least in part on an acoustic model and the updated feature vectors.
[0146] At step 902 the user device may receive updated speaker adaptation parameters from a computing device (e.g., a server device). The updated speaker adaptation parameters received from the computing device may be based on a representation of at least some of utterances that the user device transmitted to the computing device. The representation of at least some of the utterances may include or consist of at least some of the feature vectors.
[0147] At step 904, the user device may incorporate the updated speaker adaptation parameters into the ASR module.
[0148] In some implementations, the feature-space speaker adaptation parameters may be updated more frequently than the updated speaker adaptation parameters are incorporated into the ASR module. For instance, the user device may be able to continuously or from time to time update the feature-space speaker adaptation parameters without input from the computing device. However, the computing device may transmit the updated speaker adaptation parameters (e.g., a new acoustic model) to the user device for incorporation into the ASR module less regularly. Particularly, the computing device may transmit a new acoustic model to the user device once the computing device has received and processed a sufficient number of feature vectors from the user device to form the new acoustic model. The updated speaker adaptation parameters may include an updated acoustic model, and incorporating the updated speaker adaptation parameters into the ASR module may involve replacing the acoustic model with the updated acoustic model. Alternatively or additionally, the updated speaker adaptation parameters may include new feature-space speaker adaptation parameters, and incorporating the updated speaker adaptation parameters into the ASR module may involve replacing the updated feature-space speaker adaptation parameters with the new feature-space speaker adaptation parameters.
[0149] In some embodiments, the feature-space speaker adaptation parameters may include a first diagonal matrix, and updating the feature vectors based on the feature-space speaker adaptation parameters may involve applying the first diagonal matrix to the feature vectors. For instance the feature vectors may be multiplied by the first diagonal matrix. Further, the updated feature-space speaker adaptation parameters may include a second diagonal matrix, and the second diagonal matrix may be applied to a first subsequently-received feature vector. For instance, the first subsequently-received feature vector may be multiplied by the second diagonal matrix.
[0150] Also, the new feature-space speaker adaptation parameters may include a first non-diagonal matrix, and replacing the updated feature-space speaker adaptation parameters with the new feature-space speaker adaptation parameters may involve replacing the second diagonal matrix with the first non-diagonal matrix. Then, after replacing the second diagonal matrix with the first non-diagonal matrix, the non-diagonal matrix may be applied to a second subsequently-received feature vector, and the non-diagonal matrix may be updated based on the second subsequently-received feature vector.
7. Conclusion
[0151] The above detailed description describes various features and functions of the disclosed systems, devices, and methods with reference to the accompanying figures. In the figures, similar symbols typically identify similar components, unless context indicates otherwise. The illustrative embodiments described in the detailed description, figures, and claims are not meant to be limiting. Other embodiments can be utilized, and other changes can be made, without departing from the spirit or scope of the subject matter presented herein. It will be readily understood that the aspects of the present disclosure, as generally described herein, and illustrated in the figures, can be arranged, substituted, combined, separated, and designed in a wide variety of different configurations, all of which are explicitly contemplated herein.
[0152] In situations in which the systems discussed here collect personal information about users, or may make use of personal information, the users may be provided with an opportunity to control whether programs or features collect user information (e.g., information about a user's social network, social actions or activities, profession, a user's preferences, or a user's current location), or to control whether and/or how to receive content from the system that may be more relevant to the user. In addition, certain data may be treated in one or more ways before it is stored or used, so that personally identifiable information is removed. For example, a user's identity may be treated so that no personally identifiable information can be determined for the user, or a user's geographic location may be generalized where location information is obtained (such as to a city, ZIP code, or state level), so that a particular location of a user cannot be determined. Thus, the user may have control over how information is collected about the user and used by the system.
[0153] With respect to any or all of the message flow diagrams, scenarios, and flow charts in the figures and as discussed herein, each step, block and/or communication may represent a processing of information and/or a transmission of information in accordance with example embodiments. Alternative embodiments are included within the scope of these example embodiments. In these alternative embodiments, for example, functions described as steps, blocks, transmissions, communications, requests, responses, and/or messages may be executed out of order from that shown or discussed, including in substantially concurrent or in reverse order, depending on the functionality involved. Further, more or fewer steps, blocks and/or functions may be used with any of the message flow diagrams, scenarios, and flow charts discussed herein, and these message flow diagrams, scenarios, and flow charts may be combined with one another, in part or in whole.
[0154] A step or block that represents a processing of information may correspond to circuitry that can be configured to perform the specific logical functions of a herein-described method or technique. Alternatively or additionally, a step or block that represents a processing of information may correspond to a module, a segment, or a portion of program code (including related data). The program code may include one or more instructions executable by a processor for implementing specific logical functions or actions in the method or technique. The program code and/or related data may be stored on any type of computer-readable medium, such as a storage device, including a disk drive, a hard drive, or other storage media.
[0155] The computer-readable medium may also include non-transitory computer-readable storage media such as computer-readable media that stores data for short periods of time like register memory, processor cache, and/or random access memory (RAM). The computer-readable media may also include non-transitory computer-readable storage media that stores program code and/or data for longer periods of time, such as secondary or persistent long term storage, like read only memory (ROM), optical or magnetic disks, and/or compact-disc read only memory (CD-ROM), for example. The computer-readable media may also be any other volatile or non-volatile storage systems. A computer-readable medium may be considered a computer-readable storage medium, for example, or a tangible storage device.
[0156] Moreover, a step or block that represents one or more information transmissions may correspond to information transmissions between software and/or hardware modules in the same physical device. However, other information transmissions may be between software modules and/or hardware modules in different physical devices.
[0157] While various aspects and embodiments have been disclosed herein, other aspects and embodiments will be apparent to those skilled in the art. The various aspects and embodiments disclosed herein are for purposes of illustration and are not intended to be limiting, with the true scope and spirit being indicated by the following claims.
Claims
What is claimed is:
1. A method comprising:
performing, by a user device, automatic speech recognition (ASR) on received utterances, wherein performing the ASR includes: generating feature vectors based on the utterances, updating the feature vectors based on feature-space speaker adaptation parameters, transcribing the utterances to text strings, wherein the transcriptions are based at least in part on an acoustic model and the updated feature vectors, and updating the feature-space speaker adaptation parameters based on the feature vectors;
transmitting, by the user device, a representation of at least some of the utterances to a computing device for development of an updated acoustic model;
after transmitting the representation, receiving, by the user device, the updated acoustic model from the computing device, wherein the updated acoustic model is based on the representation; and
replacing, by the user device, the acoustic model with the updated acoustic model, wherein the feature-space speaker adaptation parameters are updated more frequently than the acoustic model is updated, and wherein the acoustic model is updated when the computing device has received a threshold extent of the representations from the user device.
2. The method of claim 1, further comprising:
receiving, by the user device, new feature-space speaker adaptation parameters; and
replacing, by the user device, the updated feature-space speaker adaptation parameters with the new feature-space speaker adaptation parameters.
3. The method of claim 2, wherein the feature-space speaker adaptation parameters include a first diagonal matrix, wherein updating the feature vectors based on the feature-space speaker adaptation parameters comprises applying the first diagonal matrix to the feature vectors, and wherein the updated feature-space speaker adaptation parameters include a second diagonal matrix, the method further comprising:
applying the second diagonal matrix to a first subsequently-received feature vector.
4. The method of claim 3, wherein the new feature-space speaker adaptation parameters include a first non-diagonal matrix, and wherein replacing the updated feature-space speaker adaptation parameters with the new feature-space speaker adaptation parameters comprises replacing the second diagonal matrix with the first non-diagonal matrix.
5. The method of claim 4, further comprising:
after replacing the second diagonal matrix with the first non-diagonal matrix, applying the non-diagonal matrix to a second subsequently-received feature vector; and
updating the non-diagonal matrix based on the second subsequently-received feature vector.
6. The method of claim 1, wherein transmitting the representation to the computing device comprises transmitting the at least some of the feature vectors to the computing device.
7. An article of manufacture including a non-transitory computer-readable storage medium, having stored thereon program instructions that, upon execution by a user device, cause the user device to perform operations comprising:
performing automatic speech recognition (ASR) on received utterances, wherein performing the ASR includes: generating feature vectors based on the utterances, updating the feature vectors based on feature-space speaker adaptation parameters, transcribing the utterances to text strings, wherein the transcriptions are based at least in part on an acoustic model and the updated feature vectors, and updating the feature-space speaker adaptation parameters based on the feature vectors;
transmitting a representation of at least some of the utterances to a computing device for development of an updated acoustic model;
after transmitting the representation, receiving the updated acoustic model from the computing device, wherein the updated acoustic model is based on the representation; and
replacing, by the user device, the acoustic model with the updated acoustic model, wherein the feature-space speaker adaptation parameters are updated more frequently than the acoustic model is updated, and wherein the acoustic model is updated when the computing device has received a threshold extent of the representations from the user device.
8. The article of manufacture of claim 7, wherein the operations further comprise:
receiving new feature-space speaker adaptation parameters; and
replacing the updated feature-space speaker adaptation parameters with the new feature-space speaker adaptation parameters.
9. The article of manufacture of claim 8, wherein the feature-space speaker adaptation parameters include a first diagonal matrix, wherein updating the feature vectors based on the feature-space speaker adaptation parameters comprises applying the first diagonal matrix to the feature vectors, and wherein the updated feature-space speaker adaptation parameters include a second diagonal matrix, the operations further comprising:
applying the second diagonal matrix to a first subsequently-received feature vector.
10. The article of manufacture of claim 9, wherein the new feature-space speaker adaptation parameters include a first non-diagonal matrix, and wherein replacing the updated feature-space speaker adaptation parameters with the new feature-space speaker adaptation parameters comprises replacing the second diagonal matrix with the first non-diagonal matrix.
11. The article of manufacture of claim 10, the operations further comprising:
after replacing the second diagonal matrix with the first non-diagonal matrix, applying the non-diagonal matrix to a second subsequently-received feature vector; and
updating the non-diagonal matrix based on the second subsequently-received feature vector.
12. The article of manufacture of claim 7, wherein transmitting the representation to the computing device comprises transmitting the at least some of the feature vectors to the computing device.
13. A computing device comprising:
an automatic speech recognition (ASR) module configured to generate feature vectors based on received utterances, update the feature vectors based on feature-space speaker adaptation parameters, obtain transcriptions of the utterances to text strings, wherein the transcriptions are based at least in part on an acoustic model and the updated feature vectors, and update the feature-space speaker adaptation parameters based on the feature vectors; and
a communication module configured to transmit a representation of at least some of the utterances to a server device for development of an updated acoustic model, after transmitting the representation, receive the updated acoustic model from the server device, wherein the updated acoustic model is based on the representation, and
wherein the ASR module is further configured to replace the acoustic model with the updated acoustic model, wherein the feature-space speaker adaptation parameters are updated more frequently than the acoustic model is updated, and wherein the acoustic model is updated when the server device has received a threshold extent of the representations from the computing device.
14. The computing device of claim 13, wherein the communication module is further configured to receive new feature-space speaker adaptation parameters, and wherein the ASR module is further configured to replace the updated feature-space speaker adaptation parameters with the new feature-space speaker adaptation parameters.
15. The computing device of claim 14, wherein the feature-space speaker adaptation parameters include a first diagonal matrix, wherein updating the feature vectors based on the feature-space speaker adaptation parameters comprises applying the first diagonal matrix to the feature vectors, wherein the updated feature-space speaker adaptation parameters include a second diagonal matrix, and wherein the ASR module is further configured to apply the second diagonal matrix to a first subsequently-received feature vector.
16. The computing device of claim 15, wherein the new feature-space speaker adaptation parameters include a first non-diagonal matrix, and wherein replacing the updated feature-space speaker adaptation parameters with the new feature-space speaker adaptation parameters comprises replacing the second diagonal matrix with the first non-diagonal matrix.
17. The computing device of claim 16, wherein the ASR module is further configured to, after replacing the second diagonal matrix with the first non-diagonal matrix, apply the non-diagonal matrix to a second subsequently-received feature vector, and update the non-diagonal matrix based on the second subsequently-received feature vector.
18. A system comprising:
a user device, including an automatic speech recognition (ASR) module configured to perform ASR, the ASR including: generating feature vectors based on the utterances, updating the feature vectors based on feature-space speaker adaptation parameters, transcribing the utterances to text strings, wherein the transcriptions are based at least in part on an acoustic model and the updated feature vectors, and updating the feature-space speaker adaptation parameters based on the feature vectors; and
a computing device that receives representations of one or more of the utterances transcribed by the user device for development of an updated acoustic model, and after receiving the representations transmits the updated acoustic model, based on the representations, to the user device, wherein the user device replaces the acoustic model with the updated acoustic model, wherein the feature-space speaker adaptation parameters are updated more frequently than the acoustic model is updated, and wherein the acoustic model is updated when the computing device has received a threshold extent of the representations from the user device.
19. The system of claim 18, wherein the user device is further configured to receive new feature-space speaker adaptation parameters, and wherein the ASR module is further configured to replace the updated feature-space speaker adaptation parameters with the new feature-space speaker adaptation parameters.
20. The system of claim 19, wherein the feature-space speaker adaptation parameters include a first diagonal matrix, wherein updating the feature vectors based on the feature-space speaker adaptation parameters comprises applying the first diagonal matrix to the feature vectors, wherein the updated feature-space speaker adaptation parameters include a second diagonal matrix, and wherein the ASR module is further configured to apply the second diagonal matrix to a first subsequently-received feature vector.
21. The system of claim 20, wherein the new feature-space speaker adaptation parameters include a first non-diagonal matrix, and wherein replacing the updated feature-space speaker adaptation parameters with the new feature-space speaker adaptation parameters comprises replacing the second diagonal matrix with the first non-diagonal matrix.
22. The system of claim 21, wherein the ASR module is further configured to, after replacing the second diagonal matrix with the first non-diagonal matrix, apply the non-diagonal matrix to a second subsequently-received feature vector, and update the non-diagonal matrix based on the second subsequently-received feature vector.
Patent Citations (123)
| Patent | Date | Inventor | Cited By |
|---|---|---|---|
| US4741036(A) | 1988-04-01 | Bahl et al. | Applicant |
| US4956865(A) | 1990-09-01 | Lennig et al. | Applicant |
| US5086385(A) | 1992-02-01 | Launey et al. | Applicant |
| US5233681(A) | 1993-08-01 | Bahl et al. | Applicant |
| US5257314(A) | 1993-10-01 | Kimura et al. | Applicant |
| US5278942(A) | 1994-01-01 | Bahl et al. | Examiner |
| US5428707(A) | 1995-06-01 | Gould et al. | Applicant |
| US5606644(A) | 1997-02-01 | Chou et al. | Applicant |
| US5625748(A) | 1997-04-01 | McDonough et al. | Applicant |
| US5675704(A) | 1997-10-01 | Juang et al. | Applicant |
| US5799276(A) | 1998-08-01 | Komissarchik et al. | Applicant |
| US5864810(A) | 1999-01-01 | Digalakis et al. | Examiner |
| US5895447(A) | 1999-04-01 | Ittycheriah et al. | Applicant |
| US6052682(A) | 2000-04-01 | Miller et al. | Applicant |
| US6064959(A) | 2000-05-01 | Young et al. | Applicant |
| US6073097(A) | 2000-06-01 | Gould et al. | Applicant |
| US6085160(A) | 2000-07-01 | D'hoore et al. | Applicant |
| US6112175(A) | 2000-08-01 | Chengalvarayan | Examiner |
| US6138095(A) | 2000-10-01 | Gupta et al. | Applicant |
| US6151575(A) | 2000-11-01 | Newman et al. | Applicant |
| US6167377(A) | 2000-12-01 | Gillick et al. | Applicant |
| US6185531(B1) | 2001-02-01 | Schwartz et al. | Applicant |
| US6223159(B1) | 2001-04-01 | Ishii | Examiner |
| US6224636(B1) | 2001-05-01 | Wegmann et al. | Applicant |
| US6236963(B1) | 2001-05-01 | Naito et al. | Applicant |
| US6243679(B1) | 2001-06-01 | Mohri et al. | Applicant |
| US6253181(B1) | 2001-06-01 | Junqua | Examiner |
| US6263308(B1) | 2001-07-01 | Heckerman et al. | Examiner |
| US6332122(B1) | 2001-12-01 | Ortega et al. | Applicant |
| US6334102(B1) | 2001-12-01 | Lewis et al. | Applicant |
| US6418411(B1) | 2002-07-01 | Gong | Applicant |
| US6456975(B1) | 2002-09-01 | Chang | Examiner |
| US6463413(B1) | 2002-10-01 | Applebaum et al. | Examiner |
| US6470314(B1) | 2002-10-01 | Dharanipragada et al. | Examiner |
| US6519562(B1) | 2003-02-01 | Phillips et al. | Applicant |
| US6567776(B1) | 2003-05-01 | Chang et al. | Applicant |
| US6785654(B2) | 2004-08-01 | Cyr et al. | Examiner |
| US6804647(B1) | 2004-10-01 | Heck et al. | Applicant |
| US6865528(B1) | 2005-03-01 | Huang et al. | Applicant |
| US6901364(B2) | 2005-05-01 | Nguyen et al. | Applicant |
| US6915259(B2) | 2005-07-01 | Rigazio et al. | Examiner |
| US6941264(B2) | 2005-09-01 | Konopka et al. | Examiner |
| US6999926(B2) | 2006-02-01 | Yuk et al. | Examiner |
| US7062436(B1) | 2006-06-01 | Odell et al. | Applicant |
| US7076428(B2) | 2006-07-01 | Anastasakos et al. | Applicant |
| US7092888(B1) | 2006-08-01 | McCarthy et al. | Applicant |
| US7269555(B2) | 2007-09-01 | Yuk et al. | Examiner |
| US7328154(B2) | 2008-02-01 | Mutel et al. | Examiner |
| US7337115(B2) | 2008-02-01 | Liu et al. | Applicant |
| US7418386(B2) | 2008-08-01 | Lai et al. | Applicant |
| US7437296(B2) | 2008-10-01 | Inoue et al. | Applicant |
| US7499857(B2) | 2009-03-01 | Gunawardana | Examiner |
| US7584102(B2) | 2009-09-01 | Hwang et al. | Applicant |
| US7668718(B2) | 2010-02-01 | Kahn et al. | Applicant |
| US7680659(B2) | 2010-03-01 | Gao et al. | Applicant |
| US7725316(B2) | 2010-05-01 | Chengalvarayan et al. | Examiner |
| US7729909(B2) | 2010-06-01 | Rigazio et al. | Examiner |
| US7756708(B2) | 2010-07-01 | Cohen et al. | Applicant |
| US7822603(B1) | 2010-10-01 | Parthasarathy et al. | Examiner |
| US7822605(B2) | 2010-10-01 | Zigel et al. | Applicant |
| US7912447(B2) | 2011-03-01 | Bennett, III et al. | Applicant |
| US8014591(B2) | 2011-09-01 | Baker | Applicant |
| US8032537(B2) | 2011-10-01 | Consul et al. | Applicant |
| US8050908(B2) | 2011-11-01 | Mohri et al. | Applicant |
| US8082147(B2) | 2011-12-01 | Parthasarathy et al. | Examiner |
| US8095356(B2) | 2012-01-01 | Kempe et al. | Applicant |
| US8260615(B1) | 2012-09-01 | Nakajima et al. | Applicant |
| US8296138(B2) | 2012-10-01 | Parthasarathy et al. | Examiner |
| US8306819(B2) | 2012-11-01 | Liu et al. | Examiner |
| US8346551(B2) | 2013-01-01 | Herbig et al. | Examiner |
| US2002/0049600(A1) | 2002-04-01 | L'Esperance et al. | Applicant |
| US2002/0055844(A1) | 2002-05-01 | L'Esperance et al. | Applicant |
| US2002/0087987(A1) | 2002-07-01 | Dudkiewicz et al. | Applicant |
| US2002/0123891(A1) | 2002-09-01 | Epstein | Applicant |
| US2002/0138274(A1) | 2002-09-01 | Sharma et al. | Examiner |
| US2003/0050780(A1) | 2003-03-01 | Rigazio et al. | Examiner |
| US2004/0088162(A1) | 2004-05-01 | He et al. | Applicant |
| US2004/0093210(A1) | 2004-05-01 | Toyama | Examiner |
| US2004/0107099(A1) | 2004-06-01 | Charlet | Applicant |
| US2004/0230424(A1) | 2004-11-01 | Gunawardana | Examiner |
| US2004/0249628(A1) | 2004-12-01 | Chelba et al. | Applicant |
| US2005/0028208(A1) | 2005-02-01 | Ellis et al. | Applicant |
| US2005/0044053(A1) | 2005-02-01 | Moreno et al. | Applicant |
| US2005/0131688(A1) | 2005-06-01 | Goronzy et al. | Applicant |
| US2006/0009972(A1) | 2006-01-01 | Yuk et al. | Examiner |
| US2006/0074664(A1) | 2006-04-01 | Lam et al. | Applicant |
| US2006/0132326(A1) | 2006-06-01 | Fang et al. | Applicant |
| US2006/0235684(A1) | 2006-10-01 | Chang | Examiner |
| US2007/0005355(A1) | 2007-01-01 | Tian et al. | Examiner |
| US2007/0129943(A1) | 2007-06-01 | Lei et al. | Examiner |
| US2008/0010057(A1) | 2008-01-01 | Chengalvarayan et al. | Examiner |
| US2008/0091423(A1) | 2008-04-01 | Roy et al. | Applicant |
| US2008/0120112(A1) | 2008-05-01 | Jordan et al. | Applicant |
| US2008/0177545(A1) | 2008-07-01 | Li et al. | Applicant |
| US2008/0195389(A1) | 2008-08-01 | Zhang et al. | Applicant |
| US2008/0215311(A1) | 2008-09-01 | Chelba et al. | Applicant |
| US2008/0270449(A1) | 2008-10-01 | Gossweiler et al. | Applicant |
| US2008/0312926(A1) | 2008-12-01 | Vair et al. | Applicant |
| US2009/0024390(A1) | 2009-01-01 | Deshmukh et al. | Examiner |
| US2009/0063483(A1) | 2009-03-01 | Chen et al. | Applicant |
| US2009/0112592(A1) | 2009-04-01 | Candelore | Applicant |
| US2010/0057453(A1) | 2010-03-01 | Valsan | Applicant |
| US2010/0076765(A1) | 2010-03-01 | Zweig et al. | Examiner |
| US2010/0114572(A1) | 2010-05-01 | Tani et al. | Applicant |
| US2010/0138222(A1) | 2010-06-01 | Herbig et al. | Examiner |
| US2010/0169094(A1) | 2010-07-01 | Akamine et al. | Applicant |
| US2010/0228548(A1) | 2010-09-01 | Liu et al. | Examiner |
| US2010/0296654(A1) | 2010-11-01 | Wilson et al. | Applicant |
| US2010/0312557(A1) | 2010-12-01 | Strom et al. | Applicant |
| US2010/0332235(A1) | 2010-12-01 | David | Applicant |
| US2011/0029307(A1) | 2011-02-01 | Parthasarathy et al. | Examiner |
| US2011/0054895(A1) | 2011-03-01 | Phillips et al. | Applicant |
| US2011/0055256(A1) | 2011-03-01 | Phillips et al. | Applicant |
| US2011/0067059(A1) | 2011-03-01 | Johnston et al. | Applicant |
| US2011/0134320(A1) | 2011-06-01 | Daly | Applicant |
| US2011/0213613(A1) | 2011-09-01 | Cohen et al. | Applicant |
| US2011/0295590(A1) | 2011-12-01 | Lloyd et al. | Examiner |
| US2011/0313775(A1) | 2011-12-01 | Laligand et al. | Applicant |
| US2012/0022869(A1) | 2012-01-01 | Lloyd et al. | Examiner |
| US2012/0072212(A1) | 2012-03-01 | Parthasarathy et al. | Examiner |
| US2012/0078621(A1) | 2012-03-01 | Kanevsky et al. | Applicant |
| US2012/0109651(A1) | 2012-05-01 | Chen | Applicant |
| US2012/0150539(A1) | 2012-06-01 | Jeon et al. | Applicant |
Non-Patent Literature (68)
- Shilei Zhang; Yong Qin; , “Model dimensionality selection in bilinear transformation for feature space MLLR rapid speaker adaptation,” Acoustics, Speech and Signal Processing (ICASSP), 2012 IEEE International Conference on , vol., no., pp. 4353-4356, Mar. 25-30, 2012 doi: 10.1109/ICASSP.2012.6288883.Examiner
- Varadarajan, B.; Povey, D.; Chu, S.M.; , “Quick fmllr for speaker adaptation in speech recognition,” Acoustics, Speech and Signal Processing, 2008. ICASSP 2008. IEEE International Conference on , vol., no., pp. 4297-4300, Mar. 31, 2008-Apr. 4, 2008.Examiner
- Y. Li, H. Erdogan, Y. Gao, and E. Marcheret, “Incremental online feature space MLLR adaptation for telephony speech recognition,” Proc Interspeech, pp. 1417-1420, 2002.Examiner
- Collins and Koo, “Discriminative Reranking for Natural Language Parsing,” Computational Linguistics, vol. 31, No. 1, pp. 25-70 (Mar. 2005).Applicant
- McDonald et al., “Distributed Training Strategies for the Structure Perceptron,” Proceeding HLT '10 Human Language Technologies: The 2010 Annual Conference for the North American Chapter of the Association for Computational Linguistics, pp. 1-9 (2010).Applicant
- Hall et al., “MapReduce/Bigtable for Distributed Optimization,” Neural Information Processing Systesm Workshop on Leaning on Cores, Clusters, and Clouds, pp. 1-7 (2010).Applicant
- Saraclar et al, “Joint Discriminative Language Modeling and Utterance Classification,” IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP'05) vol. 1, pp. 561-564 (2005).Applicant
- Roark et al., “Discriminative n-gram language modeling,” Computer Speech and Language, vol. 21, pp. 373-392 (2007).Applicant
- Mohri et al., “Speech Recognition with Weighted Finite-State Transducers,” Springer Handbook on Speech Processing and Speech Communication, Part E, Chapter 28, pp. 1-31 (2008).Applicant
- Mohri, Mehryar, “Weighted Automata Algorithms,” Handbook of Weighted Automata. Monographs in Theoretical Computer Science, pp. 213-254, Springer (2009).Applicant
- Goodman, Joshua T., “A Bit of Progress in Language Modeling Extended Version,” Computers & Electrical Engineering, vol. 37, Issue 4, pp. 559-569 (Jul. 2011).Applicant
- Roark et al , “Discriminative Language Modeling with Conditional Random Fields and the Perceptron Algorithm,” ACL'04 Proceedings of the 42nd Annual Meeting on Association for Computational Linguistics, pp. 1-8 (2004).Applicant
- Rabiner et al., “Introduction to Digital Speech Processing,” Foundations and Trends in Signal Processing, vol. 1, Nos. 1-2, pp. 1-194 (2007).Applicant
- Mohri et al., “Weighted Finite-State Transducers in Speech Recognition,” Department of Computer & Information Science, Departmental Papers (CIS), University of Pennsylvania, pp. 1-13 (May 14, 2004), www.cs.nyu.edu/˜mohri/pub/csl01.pdf.Applicant
- Rosell, Magnus “An Introduction to Front-End Processing and Acoustic Features for Automatic Speech Recognition,” pp. 1-10 (Jan. 17, 2006) Swedish national graduate school of language technology “GSLT”. www.csc.kth.se/˜rosell/courses/rosell—acoustic—features.pdf.Applicant
- Collins, Michael, “Discriminative Reranking for Natural Language Parsing,” Proc 17th International Conf on Machine Learning, vol. 31, Issue: 1, Publisher: Morgan Kaufmann, San Francisco, CA, pp. 175-182 (2000).Applicant
- “A Brief Introduction to Perceptrons,” pp. 1-2, www.cs.utsa.edu/˜bylander/cs5233/perceptron.pdf (Accessed Apr. 20, 2012).Applicant
- Glass, James and Zue, Victor, Lecture 7, Pattern Classification, pp. 1-33, 6.345 Automatic Speech Recognition, Spring 2003. (Massachusetts Institute of Technology: MIT OpenCourseWare), License: Creative Commons BY-NC-SA, http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-345-automatic-speech-recognition-spring-2003/lecture-notes/lecture7.pdf (Accessed Apr. 20, 2012).Applicant
- Glass, James and Zue, Victor, Lecture 9, Dynamic Time Warping & Search, pp. 1-33, 6.345 Automatic Speech Recognition, Spring 2003. (Massachusetts Institute of Technology: MIT OpenCourseWare), License: Creative Commons BY-NC-SA, http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-345-automatic-speech-recognition-spring-2003/lecture-notes/lecture9.pdf (Accessed Apr. 20, 2012).Applicant
- Glass, James and Zue, Victor, Lecture 10, Hidden Markov Modelling, pp. 1-33, 6.345 Automatic Speech Recognition, Spring 2003. (Massachusetts Institute of Technology: MIT OpenCourseWare), License: Creative Commons BY-NC-SA, http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-345-automatic-speech-recognition-spring-2003/lecture-notes/lecture10.pdf (Accessed Apr. 20, 2012).Applicant
- Glass, James and Zue, Victor, Lecture 11-12, Language Modelling for Speech Recognition, pp. 1-47, 6.345 Automatic Speech Recognition, Spring 2003. (Massachusetts Institute of Technology: MIT OpenCourseWare), License: Creative Commons BY-NC-SA, http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-345-automatic-speech-recognition-spring-2003/lecture-notes/lecture1112.pdf (Accessed Apr. 20, 2012).Applicant
- Glass, James and Zue, Victor, Lecture 13, A Practical Introduction to Graphical Models and their use in ASR, pp. 1-37, 6.345 Automatic Speech Recognition, Spring 2003. (Massachusetts Institute of Technology: MIT OpenCourseWare), License: Creative Commons BY-NC-SA, http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-345-automatic-speech-recognition-spring-2003/lecture-notes/lecture13.pdf (Accessed Apr. 20, 2012).Applicant
- Glass, James and Zue, Victor, Guest Lecture 14 by Rita Singh, Part I: Designing HMM-based ASR systems, pp. 1-68, 6.345 Automatic Speech Recognition, Spring 2003. (Massachusetts Institute of Technology: MIT OpenCourseWare), License: Creative Commons BY-NC-SA, http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-345-automatic-speech-recognition-spring-2003/lecture-notes/lecture14.pdf (Accessed Apr. 20, 2012).Applicant
- Glass, James and Zue, Victor, Guest Lecture 15 by Rita Singh, Part I: Designing HMM-based ASR systems, Part II: Training continueous density HMMS, pp. 1-66, 6.345 Automatic Speech Recognition, Spring 2003. (Massachusetts Institute of Technology: MIT OpenCourseWare), License: Creative Commons BY-NC-SA, http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-345-automatic-speech-recognition-spring-2003/lecture-notes/lecture15.pdf (Accessed Apr. 20, 2012).Applicant
- Glass, James and Zue, Victor, Lecture 17, Finite-State Techniques for Speech Recognition, pp. 1-34, 6.345 Automatic Speech Recognition, Spring 2003. (Massachusetts Institute of Technology: MIT OpenCourseWare), License: Creative Commons BY-NC-SA, http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-345-automatic-speech-recognition-spring-2003/lecture-notes/lecture 17.pdf (Accessed Apr. 20, 2012).Applicant
- Glass, James and Zue, Victor, Lecture 18, ASR for Spoken-Dialogue Systems, pp. 1-34, 6.345 Automatic Speech Recognition, Spring 2003. (Massachusetts Institute of Technology: MIT OpenCourseWare), License: Creative Commons BY-NC-SA, http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-345-automatic-speech-recognition-spring-2003/lecture-notes/lecture18.pdf (Accessed Apr. 20, 2012).Applicant
- Office Action for U.S. Appl. No. 13/461,093 mailed Aug. 2, 2012, 18 pages.Applicant
- Office Action for U.S. Appl. No. 13/456,671 mailed Jun. 13, 2012, 16 pages.Applicant
- Notice of Allowance for U.S. Appl. No. 13/456,671 mailed Oct. 9, 2012, 8 pages.Applicant
- Breslin, C. et al., “Prior Information for Rapid Speaker Adaptation,” In Proceedings of Interspeech, pp. 1644-1647 (2010).Applicant
- Ferras, Marc et al., “Constrained MLLR for Speaker Recognition,” IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2007), vol. 4, pp. IV-53 to IV-56 (2007).Applicant
- Gales, M.J.F., “Maximum Likelihood Linear Transformations for HMM-based Speech Recognition,” CUED/F-INFENG/TR 291, Tech. Rep., Cambridge University Engineering Department, pp. 1-19 (May 1997).Applicant
- Hasegawa-Johnson, Mark, Lecture 2: Acoustic Features, Acoustic Model Training, and Unix Scripting, pp. 1-16 (Jan. 8, 2009).Applicant
- Leggetter, C.J. et al., “Maximum likelihood linear regression for speaker adaptation of continuous density hidden Markov models,” Computer Speech and Language, vol. 9, pp. 171-185 (1995).Applicant
- Li, Yongxin et al., “Incremental On-Line Feature Space MLLR Adaptation for Telephony Speech Recognition.” In International Conference on Spoken Language Processing, Denver, pp. 1417-1420 (2002).Applicant
- Matsoukas et al., “Improved Speaker Adaptation Using Speaker Dependent Feature Projections,” IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU'03), pp. 273-278 (2003).Applicant
- Woodland, Phil C. “Speaker Adaptation for Continuous Density HMMs: A Review.” In ISCA Tutorial and Research Workshop (ITRW) on Adaptation Methods for Speech Recognition, pp. 11-19 (Aug. 29-30, 2001).Applicant
- Reynolds, Douglas, “Gaussian mixture models.” Encyclopedia of Biometric Recognition, pp. 12-17 (2008).Applicant
- Reynolds et al., “Speaker Verification Using Adapted Gaussian Mixture Models,” Digital Signal Processing, vol. 10, pp. 19-41 (2000).Applicant
- Reynolds et al., “Robust Text-Independent Speaker Identification Using Gaussian Mixture Speaker Models,” IEEE Transactions on Speech and Audio Processing, vol. 3, No. 1, pp. 72-83 (Jan. 1995).Applicant
- Saon, George, et al., “Eliminating Inter-Speaker Variability Prior to Discriminant Transforms.” IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU'01) pp. 73-76 (2001).Applicant
- Saon, George, et al., “Linear feature space projections for speaker adaptation.” 2001 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP'01) vol. 1, pp. 325-328, (2001).Applicant
- Tran, Dat et al., “Fuzzy Gaussian Mixture Models for Speaker Recognition,” In Proceedings of the International Conference on Spoken Language Processing, pp. 759-762 (1998) Double Checking This One Out.Applicant
- Lei et al., “Robust Feature Space Adaptation for Telephony Speech Recognition.” In Proc. ICSLP, pp. 773-776 (2006).Applicant
- Office Action for U.S. Appl. No. 13/653,792 mailed Mar. 27, 2013, 26 pages.Applicant
- Allauzen, Cyril et al., “OpenFst: A General and Efficient Weighted Finite-State Transducer Library.” In Proceedings of the 12th international conference on Implementation and application of automata, 2007, pp. 11-23. Springer-Verlag.Applicant
- Allauzen, Cyril et al., “OpenFst: A General and Efficient Weighted Finite-State Transducer Library,” http://www.stringology.org/event/CIAA2007/pres/Tue2/Riley.pdf, Jul. 17, 2007, pp. 1-36.Applicant
- Bahl et al., “A Maximum Likelihood Approach to Continuous Speech Recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Mar. 1983, pp. 179-190, vol. PAMI-5, No. 2.Applicant
- Fong et al., “Detecting Word Substitutions in Text,” IEEE Transactions on Knowledge and Data Engineering, Aug. 2008, vol. 20, No. 8, pp. 1067-1076.Applicant
- Fontanari, Jose F., “Minimal models for text production and Zipf's law,” Integration of Knowledge Intensive Multi-Agent Systems, KIMAS 2005, pp. 297-300.Applicant
- Goel et al., “Minimum Bayes-Risk Automatic Speech Recognition.” Computer Speech & Language, 2000, pp. 115-135, vol. 14, No. 2.Applicant
- Joachims et al., “SVMstruct-Support Vector Machine for Complex Outputs,” Cornell University, Department of Computer Science, Aug. 14, 2008, pp. 1-6.Applicant
- Ljolje et al. “Efficient general lattice generation and rescoring.” Sixth European Conference on Speech Communication and Technology, 1999, pp. 1-4.Applicant
- Mangu et al., “Finding consensus in speech recognition: word error minimization and other applications of confusion networks,” Computer Speech & Language, 2000, pp. 373-400, vol. 14, No. 4.Applicant
- Mohri et al., “The Design Principles of a Weighted Finite-State Transducer Library,” Theoretical Computer Science, 2000, pp. 17-32, vol. 231, No. 1.Applicant
- Mohri et al., “Weighted Finite-State Transducers in Speech Recognition,” Computer Speech & Language, 2002, pp. 69-88, vol. 16, No. 1.Applicant
- Mohri, Mehryar. “Weighted Finite-State Transducer Algorithms. An Overview.” Studies in Fuzziness and Soft Computing, 2004, pp. 551-564, vol. 148.Applicant
- Mohri, M., “Edit-Distance of Weighted Automata: General Definitions and Algorithms,” International Journal of Foundations of Computer Science, 2003, pp. 957-982, vol. 14, No. 06.Applicant
- Mohri et al., “A Rational Design for a Weighted Finite-State Transducer Library.” Automata Implementation, 1998, pp. 144-158.Applicant
- OpenFst Examples, http://www.openfst.org/twiki/bin/view/FST/FstExamples, pp. 1-9.Applicant
- Post et al., “Weight pushing and binarization for fixed-grammar parsing,” In Proceedings of the 11th International Conference on Parsing Technologies Association for Computational Linguistics, 2009, pp. 89-98.Applicant
- Schluter, R. et al., “Bayes Risk Minimization Using Metric Loss Functions,” In Proceedings of the European Conference on Speech Communication and Technology, Interspeech, 2005, pp. 1449-1452.Applicant
- Stolcke, Andreas et al., “Explicit word error minimization in N-best list rescoring.” In Proc. Eurospeech, 1997, pp. 163-166, vol. 1.Applicant
- Tsochantaridis et al., “Large Margin Methods for Structured and Interdependent Output Variables,” Journal of Machine Learning Research, 2005, pp. 1453-1484, vol. 6.Applicant
- Velivelli et al., “Automatic Video Annotation by Mining Speech Transcripts,” Proceedings of the 2006 Conference on Computer Vision and Pattern Recognition Workshop, 2006, pp. 1-8.Applicant
- Xu, Haihua et al., “Minimum Bayes Risk Decoding and System Combination Based on a Recursion for Edit Distance.” Computer Speech & Language, 2011, pp. 802-828, vol. 25, No. 4.Applicant
- Office Action for U.S. Appl. No. 13/666,223 mailed Apr. 8, 2013, 7 pages.Applicant
- Final Office Action for U.S. Appl. No. 13/461,093 mailed Jan. 14, 2013, 25 pages.Applicant