Negative n-gram biasing

• US 9,691,380 B2
• Filed: 06/15/2015
• Issued: 06/27/2017
• Est. Priority Date: 06/15/2015
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Aleksic, Petar, Moreno Mengibar, Pedro J.
• Primary Examiner(s): Cordero, Marivelisse Santiago
• Assistant Examiner(s): BRINICH, STEPHEN M
• Application Number:
  US14/739,287
• Publication Number:
  US 20160365092A1
• Time in Patent Office: 743 Days
• Field of Search: 704 1- 10, 704 50-257, 704266, 704275
• CPC Class Codes: G10L 15/01 Assessment or evaluation of speech recognition systems
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams
• CPC Class Codes: G10L 2015/228 of application context

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for performing dynamic, stroke-based alignment of touch displays. In one aspect, a method includes obtaining a candidate transcription that an automated speech recognizer generates for an utterance, determining a particular context associated with the utterance, determining that a particular n-gram that is included in the candidate transcription is included among a set of undesirable n-grams that is associated with the context, adjusting a speech recognition confidence score associated with the transcription based on determining that the particular n-gram that is included in the candidate transcription is included among the set of undesirable n-grams that is associated with the context, and determining whether to provide the candidate transcription for output based at least on the adjusted speech recognition confidence score.

Claims

1. A computer-implemented method comprising:

obtaining a candidate transcription that an automated speech recognizer generates for an utterance;

determining a particular context associated with the utterance;

obtaining a context profile associated with the particular context, wherein the context profile specifies (i) one or more n-grams, and, (ii) for each of the one more n-grams, a value that reflects an extent to which the candidate transcription includes the n-gram specified in the context profile;

determining that a particular n-gram that is included in the candidate transcription is included among a set of undesirable n-grams that is associated with the context;

adjusting a speech recognition confidence score associated with the transcription based on determining that the particular n-gram that is included in the candidate transcription is included among the set of undesirable n-grams that is associated with the context, wherein adjusting the speech recognition confidence score associated with the transcription comprises multiplying a value representing the speech recognition confidence score and the value that reflects an extent to which the candidate transcription includes the n-gram specified in the context profile; and

determining whether to provide the candidate transcription for output based at least on the adjusted speech recognition confidence score.

2. The method of claim 1, wherein the set of undesirable n-grams includes common incorrect phrases associated with the particular context associated with the utterance.

3. A computer-implemented method comprising:

obtaining a candidate transcription that an automated speech recognizer generates for an utterance;

determining a particular context associated with the utterance;

determining that a particular n-gram that is included in the candidate transcription is included among a set of undesirable n-grams that is associated with the context;

adjusting a speech recognition confidence score associated with the transcription based on determining that the particular n-gram that is included in the candidate transcription is included among the set of undesirable n-grams that is associated with the context; and

determining whether to provide the candidate transcription for output based at least on the adjusted speech recognition confidence score, wherein determining a particular context associated with the utterance comprises;

determining a likely context associated with a user submitting the utterance;

comparing the likely context associated with the user to a particular context specified by the context profile; and

determining, based at least on comparing the likely context associated with the user to the particular context specified by the context profile, that the particular context specified by the context profile is associated with the utterance.

4. A computer-implemented method comprising:

obtaining a candidate transcription that an automated speech recognizer generates for an utterance;

determining a particular context associated with the utterance;

determining that a particular n-gram that is included in the candidate transcription is included among a set of undesirable n-grams that is associated with the context;

adjusting a speech recognition confidence score associated with the transcription based on determining that the particular n-gram that is included in the candidate transcription is included among the set of undesirable n-grams that is associated with the context; and

determining whether to provide the candidate transcription for output based at least on the adjusted speech recognition confidence score, the method further comprising;

receiving (i) audio data that encodes an utterance, and (ii) context data;

obtaining candidate recognition hypotheses for the utterance that an automated speech recognizer generates based on the audio data;

obtaining a context profile associated with the received context data, wherein the context profile specifies (i) one or more n-grams, and, (ii) for each of the one more n-grams, a value that reflects an extent to which a score associated with a candidate recognition hypothesis that includes the n-gram is to be biased;

for each candidate recognition hypothesis that includes an n-gram specified in the context profile, biasing the score associated with the candidate recognition hypothesis based at least on the corresponding value specified for the n-gram, wherein biasing the score associated with the candidate recognition hypothesis comprises negatively biasing the score based at least on the context profile associated with the received context data and based at least on determining that the audio data comprises audio data encoding utterances that include highly inflected language queries;

selecting, from among the candidate recognition hypotheses, a particular candidate recognition hypothesis based at least on (i) the score, for each candidate recognition hypothesis that does not include an n-gram specified in the context profile, and (ii) the biased score, for each candidate recognition hypothesis that does include an n-gram specified in the context profile; and

providing the particular candidate recognition hypothesis for output.

5. The method of claim 4, wherein the score associated with a candidate recognition hypothesis represents a speech recognition confidence score.

6. The method of claim 4, wherein biasing the score associated with the candidate recognition hypothesis comprises negatively biasing the score based at least on determining that the audio data comprises low signal-to-noise ratio audio data.

7. A system comprising:

one or more computers; and

a non-transitory computer-readable medium coupled to the one or more computers having instructions stored thereon, which, when executed by the one or more computers, cause the one or more computers to perform operations comprising;

obtaining a candidate transcription that an automated speech recognizer generates for an utterance;

determining a particular context associated with the utterance;

obtaining a context profile associated with the particular context, wherein the context profile specifies (i) one or more n-grams, and, (ii) for each of the one more n-grams, a value that reflects an extent to which the candidate transcription includes the n-gram to specified in the context profile

determining that a particular n-gram that is included in the candidate transcription is included among a set of undesirable n-grams that is associated with the context;

adjusting a speech recognition confidence score associated with the transcription based on determining that the particular n-gram that is included in the candidate transcription is included among the set of undesirable n-grams that is associated with the context, wherein adjusting the speech recognition confidence score associated with the transcription comprises multiplying a value representing the speech recognition confidence score and the value that reflects an extent to which the candidate transcription includes the n-gram specified in the context profile; and

determining whether to provide the candidate transcription for output based at least on the adjusted speech recognition confidence score.

8. A system comprising:

one or more computers; and

a non-transitory computer-readable medium coupled to the one or more computers having instructions stored thereon, which, when executed by the one or more computers, cause the one or more computers to perform operations comprising;

obtaining a candidate transcription that an automated speech recognizer generates for an utterance;

determining a particular context associated with the utterance;

determining that a particular n-gram that is included in the candidate transcription is included among a set of undesirable n-grams that is associated with the context;

adjusting a speech recognition confidence score associated with the transcription based on determining that the particular n-gram that is included in the candidate transcription is included among the set of undesirable n-grams that is associated with the context; and

determining whether to provide the candidate transcription for output based at least on the adjusted speech recognition confidence score, wherein determining a particular context associated with the utterance comprises;

determining a likely context associated with a user submitting the utterance;

comparing the likely context associated with the user to a particular context specified by the context profile; and

determining, based at least on comparing the likely context associated with the user to the particular context specified by the context profile, that the particular context specified by the context profile is associated with the utterance.

9. A system comprising:

one or more computers; and

a non-transitory computer-readable medium coupled to the one or more computers having instructions stored thereon, which, when executed by the one or more computers, cause the one or more computers to perform operations comprising;

obtaining a candidate transcription that an automated speech recognizer generates for an utterance;

determining a particular context associated with the utterance;

determining that a particular n-gram that is included in the candidate transcription is included among a set of undesirable n-grams that is associated with the context;

adjusting a speech recognition confidence score associated with the transcription based on determining that the particular n-gram that is included in the candidate transcription is included among the set of undesirable n-grams that is associated with the context; and

determining whether to provide the candidate transcription for output based at least on the adjusted speech recognition confidence score, the operations further comprising;

receiving (i) audio data that encodes an utterance, and (ii) context data;

obtaining candidate recognition hypotheses for the utterance that an automated speech recognizer generates based on the audio data;

obtaining a context profile associated with the received context data, wherein the context profile specifies (i) one or more n-grams, and, (ii) for each of the one more n-grams, a value that reflects an extent to which a score associated with a candidate recognition hypothesis that includes the n-gram is to be biased;

for each candidate recognition hypothesis that includes an n-gram specified in the context profile, biasing the score associated with the candidate recognition hypothesis based at least on the corresponding value specified for the n-gram, wherein biasing the score associated with the candidate recognition hypothesis comprises negatively biasing the score based at least on the context profile associated with the received context data and based at least on determining that the audio data comprises audio data encoding utterances that include highly inflected language queries;

selecting, from among the candidate recognition hypotheses, a particular candidate recognition hypothesis based at least on (i) the score, for each candidate recognition hypothesis that does not include an n-gram specified in the context profile, and (ii) the biased score, for each candidate recognition hypothesis that does include an n-gram specified in the context profile; and

providing the particular candidate recognition hypothesis for output.

10. The system of claim 9, wherein the score associated with a candidate recognition hypothesis represents a speech recognition confidence score.

11. The system of claim 9, wherein biasing the score associated with the candidate recognition hypothesis comprises negatively biasing the score based at least on determining that the audio data comprises low signal-to-noise ratio audio data.

12. A non-transitory computer storage device encoded with a computer program, the program comprising instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:

obtaining a candidate transcription that an automated speech recognizer generates for an utterance;

determining a particular context associated with the utterance;

obtaining a context profile associated with the particular context, wherein the context profile specifies (i) one or more n-grams, and, (ii) for each of the one more n-grams, a value that reflects an extent to which the candidate transcription includes the n-gram specified in the context profile

determining that a particular n-gram that is included in the candidate transcription is included among a set of undesirable n-grams that is associated with the context;

adjusting a speech recognition confidence score associated with the transcription based on determining that the particular n-gram that is included in the candidate transcription is included among the set of undesirable n-grams that is associated with the context, wherein adjusting the speech recognition confidence score associated with the transcription comprises multiplying a value representing the speech recognition confidence score and the value that reflects an extent to which the candidate transcription includes the n-gram specified in the context profile; and

determining whether to provide the candidate transcription for output based at least on the adjusted speech recognition confidence score.