Speech recognition with selective use of dynamic language models
A computer-implemented method for transcribing an utterance includes receiving, at a computing system, speech data that characterizes an utterance of a user. A first set of candidate transcriptions of the utterance can be generated using a static class-based language model that includes a plurality of classes that are each populated with class-based terms selected independently of the utterance or the user. The computing system can then determine whether the first set of candidate transcriptions includes class-based terms. Based on whether the first set of candidate transcriptions includes class-based terms, the computing system can determine whether to generate a dynamic class-based language model that includes at least one class that is populated with class-based terms selected based on a context associated with at least one of the utterance and the user.
1 . A computer-implemented method when executed on data processing hardware causes the data processing hardware to perform operations comprising:
receiving an audio signal characterizing an utterance captured by a computing device associated with a user;
generating, using a class-based language model, a word lattice representing a transcription hypothesis path corresponding to a respective transcription for the utterance, the transcription hypothesis path including a class identifier;
obtaining, based on the word lattice, a list of user-specific class-based terms, each user-specific class-based term in the list of user-specific class-based terms belonging to a class associated with the class identifier included in the transcription hypothesis path;
modifying the word lattice to include the list of user-specific class-based terms;
re-training the class-based language model using the modified word lattice; and
providing, using the re-trained class-based language model, a speech recognition result for the utterance.
2 . The computer-implemented method of claim 1 , wherein the list of user-specific class-based terms comprise names derived from a contact list of the user.
3 . The computer-implemented method of claim 1 , wherein the word lattice is represented by a finite state transducer.
4 . The computer-implemented method of claim 1 , wherein the class-based language model comprises an n-gram language model.
5 . The computer-implemented method of claim 4 , wherein the class-based language model assigns each term in a sequence of terms in the word lattice a respective probability based on a statistical likelihood that the term would follow an immediately preceding n−1 terms.
6 . The computer-implemented method of claim 1 , wherein the data processing hardware resides on the computing device.
7 . The computer-implemented method of claim 1 , wherein the operations further comprise, prior to generating the word lattice, processing, using an acoustic model, the audio signal to generate a set of candidate phonemes or linguistic units of the utterance.
8 . The computer-implemented method of claim 1 , wherein the operations further comprise asynchronously obtaining context data associated with the user in parallel with generating the word lattice.
9 . The computer-implemented method of claim 1 , wherein the operations further comprise:
determining a sequence of terms traversed by the transcription hypothesis path includes the class identifier; and
in response to determining the sequence of terms includes the class identifier, obtaining context data associated with the user.
10 . The computer-implemented method of claim 1 , wherein the class identifier flanks a pre-defined class-based term that occurs in the word lattice.
11 . A system comprising:
data processing hardware; and
memory hardware in communication with the data processing hardware and storing instructions that when executed on the data processing hardware cause the data processing hardware to perform operations comprising
receiving an audio signal characterizing an utterance captured by a computing device associated with a user;
generating, using a class-based language model, a word lattice representing a transcription hypothesis path corresponding to a respective transcription for the utterance, the transcription hypothesis path including a class identifier;
obtaining, based on the word lattice, a list of user-specific class-based terms, each user-specific class-based term in the list of user-specific class-based terms belonging to a class associated with the class identifier included in the transcription hypothesis path;
modifying the word lattice to include the list of user-specific class-based terms;
re-training the class-based language model using the modified word lattice; and
providing, using the re-trained class-based language model, a speech recognition result for the utterance.
12 . The system of claim 11 , wherein the list of user-specific class-based terms comprise names derived from a contact list of the user.
13 . The system of claim 11 , wherein the word lattice is represented by a finite state transducer.
14 . The system of claim 11 , wherein the class-based language model comprises an n-gram language model.
15 . The system of claim 14 , wherein the class-based language model assigns each term in a sequence of terms in the word lattice a respective probability based on a statistical likelihood that the term would follow an immediately preceding n−1 terms.
16 . The system of claim 11 , wherein the data processing hardware resides on the computing device.
17 . The system of claim 11 , wherein the operations further comprise, prior to generating the word lattice, processing, using an acoustic model, the audio signal to generate a set of candidate phonemes or linguistic units of the utterance.
18 . The system of claim 11 , wherein the operations further comprise asynchronously obtaining context data associated with the user in parallel with generating the word lattice.
19 . The system of claim 11 , wherein the operations further comprise:
determining a sequence of terms traversed by the transcription hypothesis path includes the class identifier; and
in response to determining the sequence of terms in the word lattice includes the class identifier, obtaining context data associated with the user.
20 . The system of claim 11 , wherein the class identifier flanks a pre-defined class-based term that occurs in the word lattice.