ADJUSTING LANGUAGE MODELS

• US 20150269938A1
• Filed: 06/10/2015
• Published: 09/24/2015
• Est. Priority Date: 12/30/2010
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Lloyd, Matthew I.
• Granted Patent: US 9,542,945 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/223 Execution procedure of a spoken command
• CPC Class Codes: G10L 25/12 the extracted parameters being prediction coefficients
• CPC Class Codes: G10L 25/51 for comparison or discrimination

Abstract

Methods, systems, and apparatuses, including computer programs encoded on a computer storage medium, for adjusting language models. In one aspect, a method includes accessing audio data. Information that indicates a first context is accessed, the first context being associated with the audio data. At least one term is accessed. Information that indicates a second context is accessed, the second context being associated with the term. A similarity score is determined that indicates a degree of similarity between the second context and the first context. A language model is adjusted based on the accessed term and the determined similarity score to generate an adjusted language model. Speech recognition is performed on the audio data using the adjusted language model to select one or more candidate transcriptions for a portion of the audio data.

Claims

1-20. -20. (canceled)

21. A method performed by data processing apparatus, the method comprising:

receiving audio data and context data associated with the audio data;

identifying a topic based on a comparison of the received context data with second context data;

adjusting a language model based on the identified topic to adjust a likelihood that the language model indicates for one or more terms associated with the topic;

determining a transcription of the audio data using the adjusted language model; and

outputting the transcription determined using the adjusted language model.

22. The method of claim 21, wherein identifying the topic based on the comparison of the received context data with second context data includes:

identifying the topic based on a comparison of the received context data with context data associated with a particular term that is associated with the topic.

23. The method of claim 22, wherein identifying the topic based on the comparison of the received context data with context data associated with the particular term that is associated with the topic includes:

identifying the topic based on a comparison of the received context data with context data that indicates a context at a time at which the particular term was provided by a user, the particular term being designated as one of a plurality of words related to the topic.

24. The method of claim 22, wherein adjusting the language model based on the identified topic to adjust the likelihood that the language model indicates for one or more terms associated with the topic includes:

adjusting the language model based on the identified topic to adjust a likelihood that the language model indicates for one or more terms associated with the topic that are different from the particular term.

25. The method of claim 22, comprising:

selecting, from among multiple sets of terms that are each associated with a respective topic, a particular set of terms that are associated with the identified topic; and

wherein adjusting the language model based on the identified topic to adjust the likelihood that the language model indicates for one or more terms associated with the topic, includes;

adjusting the language model based on the identified topic to adjust a likelihood that the language model indicates for each of the terms in the particular set of terms associated with the topic.

26. The method of claim 21, comprising:

determining, for each of multiple terms, a similarity score that reflects a degree of similarity between (i) context data that is associated with the respective term and (ii) the received context data; and

wherein identifying the topic based on the comparison of the received context data with second context data includes;

selecting a particular term from among the multiple terms based at least on the similarity score for the particular term; and

identifying a topic that is associated with the particular term.

27. The method of claim 26, wherein determining, for each of multiple terms, a similarity score includes:

determining, for each of multiple terms, a similarity score that is indicative of a degree of similarity between;

(i) context data indicating a time associated with the term or a geographic location associated with the term, and

(ii) the received context data indicating a time associated with the received audio data or a geographic location associated with the received audio data.

28. The method of claim 26, wherein adjusting the language model based on the identified topic to adjust the likelihood that the language model indicates for one or more terms associated with the topic includes:

adjusting the language model based on the identified topic and one or more of the determined similarity scores to adjust the likelihood that the language model indicates for one or more terms associated with the topic.

29. The method of claim 26, wherein receiving audio data includes:

receiving, at a first point in time, audio data that represents speech of a user as obtained by one of a plurality of client devices on which an account associated with the user is provided; and

wherein selecting the particular term includes;

selecting the particular term from among terms that are each associated with data obtained by one of the plurality of client devices that reflects (i) input received through a user interface of the client device before the first point in time, or (ii) content which, before the first point in time, was associated the user'"'"'s account and accessible to one or more of the plurality of client devices.

30. The method of claim 21, comprising:

accessing multiple terms that are each associated with previously-received input data;

wherein identifying the topic based on the comparison of the received context data with second context data includes;

identifying a topic based on a comparison of the received context data with context data associated with a particular one of the multiple terms that is associated with the topic; and

wherein adjusting the language model based on the identified topic to adjust the likelihood that the language model indicates for one or more terms associated with the topic includes;

adjusting a language model based on the identified topic to adjust a likelihood that the language model indicates for terms associated with the topic which include (i) one or more terms associated with previously-received input data, and (ii) one or more terms not associated with previously-received input data.

31. A non-transitory computer-readable storage device having instructions stored thereon that, when executed by a computing device, cause the computing device to perform operations comprising:

receiving audio data and context data associated with the audio data;

identifying a topic based on a comparison of the received context data with second context data;

adjusting a language model based on the identified topic to adjust a likelihood that the language model indicates for one or more terms associated with the topic;

determining a transcription of the audio data using the adjusted language model; and

outputting the transcription determined using the adjusted language model.

32. The storage device of claim 31, wherein identifying the topic based on the comparison of the received context data with second context data includes:

identifying the topic based on a comparison of the received context data with context data associated with a particular term that is associated with the topic.

33. The storage device of claim 32, wherein identifying the topic based on the comparison of the received context data with context data associated with the particular term that is associated with the topic includes:

identifying the topic based on a comparison of the received context data with context data that indicates a context at a time at which the particular term was provided by a user, the particular term being designated as one of a plurality of words related to the topic.

34. The storage device of claim 32, wherein adjusting the language model based on the identified topic to adjust the likelihood that the language model indicates for one or more terms associated with the topic includes:

adjusting the language model based on the identified topic to adjust a likelihood that the language model indicates for one or more terms associated with the topic that are different from the particular term.

35. The storage device of claim 32, wherein the operations comprise:

selecting, from among multiple sets of terms that are each associated with a respective topic, a particular set of terms that are associated with the identified topic; and

wherein adjusting the language model based on the identified topic to adjust the likelihood that the language model indicates for one or more terms associated with the topic, includes;

adjusting the language model based on the identified topic to adjust a likelihood that the language model indicates for each of the terms in the particular set of terms associated with the topic.

36. A system comprising:

one or more data processing apparatus; and

a computer-readable storage device having stored thereon instructions that, when executed by the one or more data processing apparatus, cause the one or more data processing apparatus to perform operations comprising;

receiving audio data and context data associated with the audio data;

identifying a topic based on a comparison of the received context data with second context data;

adjusting a language model based on the identified topic to adjust a likelihood that the language model indicates for one or more terms associated with the topic;

determining a transcription of the audio data using the adjusted language model; and

outputting the transcription determined using the adjusted language model.

37. The system of claim 36, wherein identifying the topic based on the comparison of the received context data with second context data includes:

identifying the topic based on a comparison of the received context data with context data associated with a particular term that is associated with the topic.

38. The system of claim 37, wherein identifying the topic based on the comparison of the received context data with context data associated with the particular term that is associated with the topic includes:

identifying the topic based on a comparison of the received context data with context data that indicates a context at a time at which the particular term was provided by a user, the particular term being designated as one of a plurality of words related to the topic.

39. The system of claim 37, wherein adjusting the language model based on the identified topic to adjust the likelihood that the language model indicates for one or more terms associated with the topic includes:

adjusting the language model based on the identified topic to adjust a likelihood that the language model indicates for one or more terms associated with the topic that are different from the particular term.

40. The system of claim 37, wherein the operations comprise:

selecting, from among multiple sets of terms that are each associated with a respective topic, a particular set of terms that are associated with the identified topic; and

wherein adjusting the language model based on the identified topic to adjust the likelihood that the language model indicates for one or more terms associated with the topic, includes;

adjusting the language model based on the identified topic to adjust a likelihood that the language model indicates for each of the terms in the particular set of terms associated with the topic.