Speech endpointing based on word comparisons
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for speech endpointing based on word comparisons are described. In one aspect, a method includes the actions of obtaining a transcription of an utterance. The actions further include determining, as a first value, a quantity of text samples in a collection of text samples that (i) include terms that match the transcription, and (ii) do not include any additional terms. The actions further include determining, as a second value, a quantity of text samples in the collection of text samples that (i) include terms that match the transcription, and (ii) include one or more additional terms. The actions further include classifying the utterance as a likely incomplete utterance or not a likely incomplete utterance based at least on comparing the first value and the second value.
1. A computer-implemented method when executed on data processing hardware causes the data processing hardware to perform operations comprising:
receiving audio data corresponding to an utterance comprising a sequence of terms;
processing, using a speech recognizer, the audio data to generate a transcription of the utterance by recognizing each term in the sequence of terms;
recording an amount of time after each term is recognized by the speech recognizer;
identifying an endpoint after a respective one of the terms recognized by the speech recognizer when the recorded amount of time after the respective one of the terms is recognized by the speech recognizer satisfies a first threshold before a subsequent term from the sequence of terms is recognized by the speech recognizer;
subsequent to identifying the endpoint after the respective one of the terms recognized by the speech recognizer, determining that the utterance is likely incomplete based on the transcription of the utterance; and
overriding the endpoint identified after the respective one of the terms recognized by the speech recognizer based on determining that the utterance is likely incomplete.
2. The method of claim 1 , wherein determining that the utterance is likely incomplete comprises:
comparing the transcription of the utterance to a first collection of text samples identified as complete utterances;
comparing the transcription of the utterance to a second collection of text samples identified as incomplete utterances; and
determining that the utterance is likely incomplete based on comparing the transcription of the utterance to the first collection of text samples identified as complete utterances and comparing the transcription of the utterance to the second collection of text samples identified as incomplete utterances.
3. The method of claim 2 , wherein comparing the transcription of the utterance to the first collection of text samples identified as complete utterances comprises:
determining a first quantity of text samples in the first collection that match the transcription of the utterance; and
determining a second quantity of text samples in the second collection that match the transcription of the utterance.
4. The method of claim 2 , wherein comparing the transcription of the utterance to the first collection of text samples identified as complete utterances comprises:
determining whether terms in each text sample in the first collection occur in a same order as terms of the transcription of the utterance; and
determining whether terms in each text sample in the second collection occur in a same order as terms of the transcription of the utterance.
5. The method of claim 1 , wherein the operations further comprise:
subsequent to identifying the endpoint after the respective one of the terms recognized by the speech recognizer, determining that the utterance is likely complete based on the transcription of the utterance; and
based on determining that the utterance is likely complete, determining to designate the endpoint identified after the respective one of the terms recognized by the speech recognizer at an end of the audio data of the utterance.
6. The method of claim 5 , wherein the operations further comprise, based on determining to designate the endpoint identified after the respective one of the terms recognized by the speech recognizer at the end of the audio data of the utterance, deactivating a microphone that detected the utterance.
7. The method of claim 1 , wherein overriding the endpoint identified after the respective one of the terms recognized by the speech recognizer comprises maintaining a microphone that detected the utterance in an active state.
8. The method of claim 7 , wherein maintaining the microphone that detected the utterance in the active state permits the speech recognizer to process additional audio data.
9. The method of claim 1 , wherein the speech recognizer resides on a user device.
10. The method of claim 1 , wherein the speech recognizer resides on a server.
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 causes the data processing hardware to perform operations comprising:
receiving audio data corresponding to an utterance comprising a sequence of terms;
processing, using a speech recognizer, the audio data to generate a transcription of the utterance by recognizing each term in the sequence of terms;
recording an amount of time after each term is recognized by the speech recognizer;
identifying an endpoint after a respective one of the terms recognized by the speech recognizer when the recorded amount of time after the respective one of the terms is recognized by the speech recognizer satisfies a first threshold before a subsequent term of the sequence of terms is recognized by the speech recognizer;
subsequent to identifying the endpoint after the respective one of the terms recognized by the speech recognizer, determining that the utterance is likely incomplete based on the transcription of the utterance; and
overriding the endpoint identified after the respective one of the terms recognized by the speech recognizer based on determining that the utterance is likely incomplete.
12. The system of claim 11 , wherein determining that the utterance is likely incomplete comprises:
comparing the transcription of the utterance to a first collection of text samples identified as complete utterances;
comparing the transcription of the utterance to a second collection of text samples identified as incomplete utterances; and
determining that the utterance is likely incomplete based on comparing the transcription of the utterance to the first collection of text samples identified as complete utterances and comparing the transcription of the utterance to the second collection of text samples identified as incomplete utterances.
13. The system of claim 12 , wherein comparing the transcription of the utterance to the first collection of text samples identified as complete utterances comprises:
determining a first quantity of text samples in the first collection that match the transcription of the utterance; and
determining a second quantity of text samples in the second collection that match the transcription of the utterance.
14. The system of claim 12 , wherein comparing the transcription of the utterance to the first collection of text samples identified as complete utterances comprises:
determining whether terms in each text sample in the first collection occur in a same order as terms of the transcription of the utterance; and
determining whether terms in each text sample in the second collection occur in a same order as terms of the transcription of the utterance.
15. The system of claim 11 , wherein the operations further comprise:
subsequent to identifying the endpoint after the respective one of the terms recognized by the speech recognizer, determining that the utterance is likely complete based on the transcription of the utterance; and
based on determining that the utterance is likely complete, determining to designate the endpoint identified after the respective one of the terms recognized by the speech recognizer at an end of the audio data of the utterance.
16. The system of claim 15 , wherein the operations further comprise, based on determining to designate the endpoint identified after the respective one of the terms recognized by the speech recognizer at the end of the audio data of the utterance, deactivating a microphone that detected the utterance.
17. The system of claim 11 , wherein overriding the endpoint identified after the respective one of the terms recognized by the speech recognizer comprises maintaining a microphone that detected the utterance in an active state.
18. The system of claim 17 , wherein maintaining the microphone that detected the utterance in the active state permits the speech recognizer to process additional audio data.
19. The system of claim 11 , wherein the speech recognizer resides on a user device.
20. The system of claim 11 , wherein the speech recognizer resides on a server.