Language model biasing modulation
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for modulating language model biasing. In some implementations, context data is received. A likely context associated with a user is determined based on at least a portion of the context data. One or more language model biasing parameters based at least on the likely context associated with the user is selected. A context confidence score associated with the likely context based on at least a portion of the context data is determined. One or more language model teasing parameters based at least on the context confidence score is adjusted. A baseline language model based at least on the one or more of the adjusted language model biasing parameters is biased. The baseline language model is provided for use by an automated speech recognizer (ASR).
1. A method comprising:
receiving at a user device associated with a user audio data corresponding to a voice query spoken by the user;
processing, by the user device, using a language model, the audio data corresponding to the voice query spoken by the user to determine a particular context associated with biasing the language model, the particular context associated with biasing the language model based on a type of words used in the voice query;
selecting by the user device, a set of biasing terms based on the particular context associated with biasing the language model; and
biasing, by the user device, using the set of biasing terms, the language model to increase a likelihood of recognizing biasing terms from the set of biasing terms in the audio data.
2. The method of claim 1 , wherein the particular context associated with biasing the language model is further based on a number of words used in the voice query.
3. The method of claim 1 , further comprising providing, by the user device, the biased language model to an automated speech recognition module, the automated speech recognition module comprising an acoustic model configured to generate readable text from speech inputs spoken by the user.
4. The method of claim 1 , wherein using the set of biasing terms to bias the language model causes the biased language model to identify certain words or terms indicating that the user is within the particular context.
5. The method of claim 1 , further comprising, after biasing the language model:
receiving, at the user device, additional audio data corresponding to an utterance spoken by the user;
determining, by the user device, that, when the additional audio data was received the particular context associated with biasing the language model is no longer applicable: and
in response to determining that the particular context associated with biasing the language model is no longer applicable, returning, by the user device, the biased language model to a baseline state.
6. The method of claim 5 , wherein determining the particular context associated with biasing the language model is no longer applicable comprises determining that the user that spoke the utterance is likely associated with another context that is inconsistent with the particular context.
7. The method of claim 5 , wherein determining the particular context associated with biasing the language model is no longer applicable comprises determining that a confidence score associated with the particular context does not satisfy a threshold.
8. The method of claim 7 , wherein the context confidence score reflects a likelihood that the particular context associated with biasing the language model is applicable.
9. The method of claim 1 , further comprising, after biasing the language model, generating, by the user device, using the biased language model, a transcription of the received audio data corresponding to the voice query spoken by the user.
10. The method of claim 1 , wherein the language model indicates scores associated with different n-gram sequences.
11. A user device comprising:
data processing hardware; and
memory hardware in communication with the data processing hardware and storing instructions that when executed on the data processing hard-ware cause the data processing hardware to perform operations comprising:
receiving audio data corresponding to a voice query spoken by a user
processing using a language model, the audio data corresponding to the voice query spoken by the user to determine a particular context associated with biasing the language model, the particular context associated with biasing the language model based on a type of words used in the voice query:
selecting a set of biasing terms based on the particular context associated with biasing the language model; and
biasing, using the set of biasing terms, the language model to increase a likelihood of recognizing biasing terms from the set of biasing terms in the audio data.
12. The user device of claim 11 , wherein the particular context associated with biasing the language model is further based on a number of words used in the voice query.
13. The user device of claim 11 , wherein the operations further comprise providing the biased language model to an automated speech recognition module, the automated speech recognition module comprising an acoustic model configured to generate readable text from speech inputs spoken by the user.
14. The user device of claim 11 , wherein using the set of biasing terms to bias the language model causes the biased language model to identify certain words or terms indicating that the user is within the particular context.
15. The user device of claim 11 , wherein the operations further comprise, after biasing the language model:
receiving additional audio data corresponding to an utterance spoken by the user;
determining that, when the additional audio data was received, the particular context associated with biasing the language model is no longer applicable; and
in response to determining that the particular context associated with biasing the language model is no longer applicable, returning the biased language model to a baseline state.
16. The user device of claim 15 , wherein determining the particular context associated with biasing the language model is no longer applicable comprises determining that the user that spoke the utterance is likely associated with another context that is inconsistent with the particular context.
17. The user device of claim 15 , wherein determining the particular context associated with biasing the language model is no longer applicable comprises determining that a confidence score associated with the particular context does not satisfy a threshold.
18. The user device of claim 17 , wherein the context confidence score reflects a likelihood that the particular context associated with biasing the language model is applicable.
19. The user device of claim 11 , wherein the operations further comprise, after biasing the language model, generating, using the biased language model, a transcription of the received audio data corresponding to the voice query spoken by the user.
20. The user device of claim 11 , wherein the language model indicates scores associated with different n-gram sequences.