History-based ASR mistake corrections

1. A computer-implemented method executed on data processing hardware causes the data processing hardware to perform operations comprising:

receiving initial audio data captured by an assistant-enabled device while the assistant-enabled device is in a sleep state, the initial audio data comprising a hotword and a previous query submitted by a user of the assistant-enabled device to the digital assistant, the hotword when detected by the assistant-enabled device causing the assistant-enabled device to wake from a sleep state and trigger a speech recognizer to perform speech recognition on at least a portion of the initial audio data that includes the previous query;

after the speech recognizer performs speech recognition on at least the portion of the initial audio data, instructing the assistant-enabled device to operate in a follow-on query mode

receiving follow-on audio data captured by the assistant-enabled device during operation of the assistant-enabled device in the follow-on query mode, the follow-on audio data corresponding to a follow-on query spoken by the user of the assistant-enabled device to the digital assistant subsequent to the user submitting the previous query to the digital assistant;

processing, using the speech recognizer, the follow-on audio data to generate multiple candidate hypotheses, each candidate hypothesis corresponding to a candidate transcription for the follow-on query and represented by a respective sequence of hypothesized terms;

for each corresponding candidate hypothesis among the multiple candidate hypotheses:

determining a corresponding similarity metric between the previous query and the corresponding candidate hypothesis, the similarity metric indicating a similarity between a topic associated with the corresponding candidate hypothesis and a topic associated with the previous query; and

obtaining a corresponding likelihood score that the speech recognizer assigned to the corresponding candidate hypothesis;

ranking the multiple candidate hypotheses based on the corresponding likelihood scores assigned to the multiple candidate hypotheses by the speech recognizer and the corresponding similarity metrics determined for each of the corresponding candidate hypotheses among the multiple candidate hypotheses; and

determining a transcription of the follow-on query spoken by the user based on the ranking of the multiple candidate hypotheses.

2. The computer-implemented method of claim 1 , wherein the operations further comprise:

determining a previous Query Based Salient Terms (QBST) vector associated with the previous query submitted by the user; and

for each of the corresponding candidate hypotheses, determining a corresponding candidate QBST vector associated with the corresponding candidate hypothesis,

wherein determining the corresponding similarity metric between the previous query and the corresponding candidate hypothesis comprises determining the corresponding similarity metric between the previous QBST vector and the corresponding candidate QBST vector based on the previous QBST vector and the corresponding candidate QBST vector.

3. The computer-implemented method of claim 2 , wherein:

the previous QBST vector indicates a respective set of salient terms associated with the previous query; and

each of the corresponding candidate QBST vectors indicates a respective set of salient terms associated with the corresponding candidate hypothesis.

4. The computer-implemented method of claim 2 , wherein the corresponding similarity metric indicates a topical drift between the previous QBST vector and the corresponding candidate QBST vector.

5. The computer-implemented method of claim 2 , wherein the corresponding similarity metric comprises a Cosine score between the previous QBST vector and the corresponding candidate QBST vector.

6. The computer-implemented method of claim 1 , wherein the presence of the hotword is absent from the follow-on audio data.

7. The computer-implemented method of claim 1 , wherein the operations further comprise:

performing query interpretation on the transcription to identify an operation specified by the follow-on query;

instructing the digital assistant to perform the operation specified by the follow-on query; and

receiving, from the digital assistant, a follow-on response indicating performance of the operation specified by the follow-on query.

8. The computer-implemented method of claim 7 , wherein the operations further comprise presenting, for output from the assistant-enabled device, the follow-on response.

9. The computer-implemented method of claim 1 , wherein the data processing hardware resides on the assistant-enabled device and executes the speech recognizer.

10. The computer-implemented method of claim 1 , wherein the data processing hardware resides on a remote server and executes the speech recognizer, the remote server in communication with the assistant-enabled device via a network.

11. The computer-implemented method of claim 1 , wherein the speech recognizer comprises an end-to-end speech recognition model.

12. The computer-implemented method of claim 1 , wherein the speech recognizer comprises an acoustic model and a language model.

13. The computer-implemented method of claim 1 , wherein the assistant-enabled device is in communication with one or more microphones configured to capture the follow-on audio data and the initial audio data corresponding to the previous query.

14. The computer-implemented method of claim 1 , wherein the assistant-enabled device comprises a battery-powered device.

15. 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 initial audio data captured by an assistant-enabled device while the assistant-enabled device is in a sleep state, the initial audio data comprising a hotword and a previous query submitted by a user of the assistant-enabled device to the digital assistant, the hotword when detected by the assistant-enabled device causing the assistant-enabled device to wake from a sleep state and trigger a speech recognizer to perform speech recognition on at least a portion of the initial audio data that includes the previous query;

after the speech recognizer performs speech recognition on at least the portion of the initial audio data, instructing the assistant-enabled device to operate in a follow-on query mode

receiving follow-on audio data captured by the assistant-enabled device during operation of the assistant-enabled device in the follow-on query mode, the follow-on audio data corresponding to a follow-on query spoken by the user of the assistant-enabled device to the digital assistant subsequent to the user submitting the previous query to the digital assistant;

processing, using the speech recognizer, the follow-on audio data to generate multiple candidate hypotheses, each candidate hypothesis corresponding to a candidate transcription for the follow-on query and represented by a respective sequence of hypothesized terms;

for each corresponding candidate hypothesis among the multiple candidate hypotheses:

determining a corresponding similarity metric between the previous query and the corresponding candidate hypothesis, the similarity metric indicating a similarity between a topic associated with the corresponding candidate hypothesis and a topic associated with the previous query; and

obtaining a corresponding likelihood score that the speech recognizer assigned to the corresponding candidate hypothesis;

ranking the multiple candidate hypotheses based on the corresponding likelihood scores assigned to the multiple candidate hypotheses by the speech recognizer and the corresponding similarity metrics determined for each of the corresponding candidate hypotheses among the multiple candidate hypotheses; and

determining a transcription of the follow-on query spoken by the user based on the ranking of the multiple candidate hypotheses.

16. The system of claim 15 , wherein the operations further comprise:

determining a previous Query Based Salient Terms (QBST) vector associated with the previous query submitted by the user; and

for each of the corresponding candidate hypotheses, determining a corresponding candidate QBST vector associated with the corresponding candidate hypothesis,

wherein determining the corresponding similarity metric between the previous query and the corresponding candidate hypothesis comprises determining the corresponding similarity metric between the previous QBST vector and the corresponding candidate QBST vector based on the previous QBST vector and the corresponding candidate QBST vector.

17. The system of claim 16 , wherein:

the previous QBST vector indicates a respective set of salient terms associated with the previous query; and

each of the corresponding candidate QBST vectors indicates a respective set of salient terms associated with the corresponding candidate hypothesis.

18. The system of claim 16 , wherein the corresponding similarity metric indicates a topical drift between the previous QBST vector and the corresponding candidate QBST vector.

19. The system of claim 15 , wherein the corresponding similarity metric comprises a Cosine score between the previous QBST vector and the corresponding candidate QBST vector.

20. The system of claim 15 , wherein the presence of the hotword is absent from the follow-on audio data.

21. The system of claim 15 , wherein the operations further comprise:

performing query interpretation on the transcription to identify an operation specified by the follow-on query;

instructing the digital assistant to perform the operation specified by the follow-on query; and

receiving, from the digital assistant, a follow-on response indicating performance of the operation specified by the follow-on query.

22. The system of claim 21 , wherein the operations further comprise presenting, for output from the assistant-enabled device, the follow-on response.

23. The system of claim 15 , wherein the data processing hardware resides on the assistant-enabled device and executes the speech recognizer.

24. The system of claim 15 , wherein the data processing hardware resides on a remote server and executes the speech recognizer, the remote server in communication with the assistant-enabled device via a network.

25. The system of claim 15 , wherein the speech recognizer comprises an end-to-end speech recognition model.

26. The system of claim 15 , wherein the speech recognizer comprises an acoustic model and a language model.

27. The system of claim 15 , wherein the assistant-enabled device is in communication with one or more microphones configured to capture the follow-on audio data and the initial audio data corresponding to the previous query.

28. The system of claim 15 , wherein the assistant-enabled device comprises a battery-powered device.