Determining context for speech recognition

[0001] This invention relates generally to mobile processor-based systems that include speech recognition capabilities. Mobile processor-based systems include devices such as handheld devices, pocket personal computers, personal digital assistants, digital cameras, laptop computers, data input devices, data collection devices, remote control units, voice recorders, and cellular telephones, to mention a few examples. Many of these devices may include speech recognition capabilities. With speech recognition, words uttered by a user may be converted to text, enabling automated transcription of speech. Spoken words may also be received as commands that enable selection and operation of the processor-based system's capabilities. In other instances, a system may provide automatic speaker identification capabilities where a speaker may be identified by comparison of his or her speech with a set of predefined models of known speakers. In a number of cases, the ability of a given device to recognize speech or identify a speaker is relatively limited. A variety of ambient conditions may adversely affect the quality of the speech recognition or speaker identification. Because the ambient conditions may change unpredictably, the elimination of ambient effects is much more difficult with mobile speech recognition platforms than with platforms where the ambient conditions are predetermined or static.

[0002] FIG. 1 depicts a speech recognition system in one embodiment of the claimed subject matter.

[0003] FIG. 2 depicts processing performed by a speech context manager in one embodiment.

[0004] FIG. 3 depicts a calendar based speech context retrieval process in one embodiment.

[0005] Embodiments of the claimed subject matter, generally, are systems which use information about events that is available from schedule data to assist in speech recognition and speaker identification tasks. One such system is depicted in FIG. 1. Referring to FIG. 1, a speech enabled mobile processor-based system 14 may be any one of a variety of mobile processor-based systems. Examples of such devices include laptop computers, pocket personal computers, personal digital assistants, cellular telephones, digital cameras, data input devices, data collection devices, appliances, and voice recorders, to mention a few examples.

[0006] The system 14 may have a personal scheduling, appointment book, calendar manager application or calendaring or scheduling component that includes the schedule of the user of the device. Such a calendar manager may use data solely internal to the device or may communicate with an external calendar or schedule database on a separate server that is part of a networking infrastructure in order to retrieve information about the user's calendar or schedule, or may combine the two approaches, by caching data previously obtained from the server in a local memory. The server 12 may be coupled to the mobile system 14 through a wireless network 18 in one embodiment of the claimed subject matter; the system may alternatively connect to the networked server using an occasionally connected wired network connection or a remote access method such as dial-up networking; or access the server using an occasional connection to an intermediate host computer which is in turn connected to a server, via a peripheral interconnect.

[0007] The ability of system 14 to recognize spoken words may be improved in a variety of environments or ambient conditions if a recognizer 22 has data that is specific to the acoustic environment of the speaker. Such data may be deduced from information available to the calendar manager 26. The calendar manager may be able to look up the room or another location scheduled for use during an event based on the identity of the user of the system, calendar data and the current time, and using the room or location information, obtain information about the acoustic characteristics of the surrounding space either from data local to the device 14 or from data on a server data store 16. Those characteristics may include the size of the room, noise sources, such as ventilation ducts or exterior windows, the specific type of each noise source, if any, and reverberation characteristics of the space. Such data may be compiled from pre-existing measurements and information relating to the specifics of a particular building, room, or office based on construction and architectural plans, or from other prior data gathering. Alternatively, in some embodiments of the claimed subject matter, the system 14 may be able to create such data and cause it to be stored on encountering a new location, or be able to update data relating to a location for which data is already available if the system can refine or improve upon the data, or if the acoustic characteristics of the space have changed.

[0008] Other information that may be leveraged to improve speech recognition of any given speaker includes the number, identities and physical locations of other nearby speakers at the time the speech recognition task is being performed. These speakers may be potential sources of interference and can be identified based on their being listed in the calendar database as participating in the same event as the user of the system 14. Data about the other speakers' location may be explicitly available as well, if for example, seating at the event is predetermined, or in another example, if the interfering speakers are seated in an auditorium and the speaker whose speech is being recognized is on a stage. In addition, the identity of nearby people may be obtained by the speech recognition system using ad hoc discovery. The system may do this by recording and analyzing the speech of persons present at the event and then by comparing the sampled speech with stored personal speech models that may be available from the server. Alternatively, the system 14 may interact with the user and allow the user to assist it with recognizing the identities of the other speakers at the event by entering their names, in one embodiment of the claimed subject matter.

[0009] The system 14 includes a speech context manager 24 that is coupled to the calendar manager/client 26, a speech recognizer 22, and a noise mitigating speech preprocessor 20.

[0010] When speech recognition is attempted by the system 14, the speech context manager 24 retrieves a current context from the server 12 in accordance with one embodiment of the claimed subject matter. Based on the size of the surrounding space, the context manager 24 adjusts the acoustic models of the recognizer 22 to account for reverberation.

[0011] This adjustment may be done in a variety of ways including using model adaptation, such as maximum likelihood linear regression to a known target. The target transformation may have been estimated in a previous encounter at that position or may be inferred from the reverberation time associated with the space. The context manager may also perform the adjustment by selecting from a set of previously trained acoustic models that match various acoustic spaces typically encountered by the user.

[0012] As another alternative, the context manager 24 may select from among feature extraction and noise reduction algorithms that are resistant to reverberation based on the size of the acoustic space. The acoustic models may also be modified to match the selected front-end noise reduction and feature extraction. Models may also be adapted based on the identity of nearby people, retrieving and loading speaker dependent acoustic models for each person, if available. Those models may be used for automatic transcription of hallway discussion in one embodiment of the claimed subject matter.

[0013] Another way that the adjustment may be done is by initializing and adapting a new acoustic model if the acoustic space has not been encountered previously. Once the location is adequately modeled, the system 14 may send the information to the server 12 to be stored in the remote data store 16 for future visitors to the same location. Furthermore, if speaker specific acoustic models such as identification or recognition models are not available for a specific speaker at the event, or if for some reason such as an illness a speaker's speech has changed significantly and permanently, the system may update the remote data store with new or changed information regarding the new or changed speakers.

[0014] Information regarding the number of participants at the event may be used by the system 14 in initializing Blind Signal Separation (BSS). BSS is an algorithm that can improve recognition accuracy in multi-talker babble noise, and is used to pre-process input to a speech recognizer. BSS works best when correctly initialized with the number of sources of babble, or speech that is to be ignored by the system. Further, it may be possible for the system to deduce the likely pitch range of speakers from the gender of the speaker if such information is available. This may then additionally improve the performance of the BSS pre-processor by assisting it in filtering certain speakers in or out at the preprocessing stage.

[0015] Another example of adaptation, based on the identity of nearby speakers derived from schedule information, the system 14 may assist the user in identifying them in meetings or events where the user may not initially know the identities of all participants. The user may then use the names of the identified speakers to interact with the system 14 and in one embodiment, indicate to the system 14 using an input device, whether their speech should be transcribed. In some embodiments, speaker identification may be further used to automatically determine whether speech for an identified speaker should be transcribed, based, for example, on the user having previously identified speakers that he or she wishes to designate as a transcription source (i.e., a person whose speech should be transcribed). This list may be stored either in the system 14 or be available from a remote source such as the server 12. In other embodiments of the claimed subject matter, the system 14 may use the names of identified speakers in combination with organizational information that also may be available from the data store 16 at server 12 and then use that information to apply pre-defined criteria such as whether a person is in a particular department of an organization, to determine whether a specific speaker at an event should be designated as a transcription source.

[0016] As still another example, based on the orientation of the system 14, the location of proximate people, and their designation as transcription sources, a microphone array controlled by preprocessor 20 may be configured to place nulls in the direction of the closest persons who are not transcription sources. Since that direction may not be highly accurate and is subject to abrupt change, this method may not supplant interferer tracking via a microphone array. However, it may provide a mechanism to place the nulls when the interferer is not speaking, thereby significantly improving performance when an interferer starts to speak.

[0017] Referring to FIG. 2, in accordance with one embodiment of the claimed subject matter, the speech context manager 24 may be a processor-based device including both a processor and storage for storing instructions to be executed on the processor. Thus, the speech context manager 24 may be software or hardware. Initially, the speech context manager 24 retrieves a current context from the server 12, as indicated in block 30. Then the context manager 24 may determine the size of the surrounding space proximate to the device 14, as indicated in block 32. The device 14 may adjust the recognizer's acoustic models to account for local reverberation, as indicated in block 34.

[0018] Then feature extraction and noise reduction algorithms may be selected based on the understanding of the local environment, as indicated in block 36. In addition, the speaker-dependent acoustic models for nearby speakers may be retrieved and loaded, as indicated in block 38. These models may be retrieved, in one embodiment, from the server 12.

[0019] As indicated in block 44, any speakers whose speech should be recognized may be identified. The microphone array preprocessor 20 may be configured, as indicated in block 46. Then speech recognition may be implemented, as indicated in block 48, having obtained the benefit of the location information.

[0020] Referring to FIG. 3, the calendaring server 12 may be implemented through software 50 that retrieves event context in accordance with one embodiment of the claimed subject matter. The software 50 may be stored in an appropriate storage on the server 12. Initially, the server 12 receives a request for context information from a system 14, including user and time information, as indicated in block 52. Once received, the server 12 obtains the schedule information from the system 14, as indicated in block 54. The location information may then be correlated to available models in the data storage 16, as indicated in block 56. Once an appropriate model is identified, the context may be transmitted to the device 14 over the wireless network, as indicated in block 58.

[0021] The present invention includes various steps. The steps of the present invention may be performed by hardware devices, or may be embodied in machine-executable instructions, which may be used to cause a general-purpose or special-purpose processor or logic circuits programmed with the instructions to perform the steps. Alternatively, the steps may be performed by a combination of hardware and software. The present invention may be provided as a computer program product that may include a machine-readable medium having stored thereon data which when accessed by a machine, may cause the machine to perform a process according to the present invention. The machine-readable medium may include, but is not limited to, floppy diskettes, optical disks, DVD-ROM disks, DVD-RAM disks, DVD-RW disks, DVDRW disks, CD-R disks, CD-RW disks, CD-ROM disks, and magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, magnet or optical cards, flash memory, or other type of media/machine-readable medium suitable for storing electronic instructions. Moreover, the present invention may also be downloaded as a computer program product, wherein the program may be transferred from a remote computer to a requesting computer by way of data signals embodied in a carrier wave or other propagation medium via a communication link (e.g., a modem or network connection).

[0022] Many of the methods are described in their most basic form but steps can be added to or deleted from any of the methods and information can be added or subtracted from any of the described messages without departing from the basic scope of the present invention. It will be apparent to those skilled in the art that many further modifications and adaptations can be made. The particular embodiments are not provided to limit the invention but to illustrate it. The scope of the present invention is not to be determined by the specific examples provided above but only by the claims below.