INDIVIDUALIZED HOTWORD DETECTION MODELS

• US 20170186433A1
• Filed: 06/29/2016
• Published: 06/29/2017
• Est. Priority Date: 07/22/2015
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Alvarez Guevara, Raziel
• Granted Patent: US 10,535,354 B2
• Time in Patent Office: Days
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/07 to the speaker
• CPC Class Codes: G10L 15/075 supervised, i.e. under machine guidance
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• CPC Class Codes: G10L 17/04 Training, enrolment or model building
• CPC Class Codes: G10L 17/06 Decision making techniques; Pattern matching strategies
• CPC Class Codes: G10L 17/08 Use of distortion metrics or a particular distance between probe pattern and reference templates
• CPC Class Codes: G10L 17/18 Artificial neural networks; Connectionist approaches
• CPC Class Codes: G10L 17/24 the user being prompted to utter a password or a predefined phrase
• CPC Class Codes: G10L 2015/0638 Interactive procedures
• CPC Class Codes: G10L 2015/088 Word spotting

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for presenting notifications in an enterprise system. In one aspect, a method include actions of obtaining enrollment acoustic data representing an enrollment utterance spoken by a user, obtaining a set of candidate acoustic data representing utterances spoken by other users, determining, for each candidate acoustic data of the set of candidate acoustic data, a similarity score that represents a similarity between the enrollment acoustic data and the candidate acoustic data, selecting a subset of candidate acoustic data from the set of candidate acoustic data based at least on the similarity scores, generating a detection model based on the subset of candidate acoustic data, and providing the detection model for use in detecting an utterance spoken by the user.

Claims

1. (canceled)

2. A computer-implemented method comprising:

during an enrollment process, prompting, by a client device, a user to speak a particular hotword, and receiving, by the client device, audio data corresponding to only a single utterance of the particular hotword by the user;

in response to receiving the audio data corresponding to only the single utterance of the particular hotword by the user, obtaining, by the client device, a personalized hotword detection model, wherein the personalized hotword detection model is trained to detect a likely utterance of the particular hotword by the user using the audio data corresponding to only the single utterance of the particular hotword by the user and without requiring the user to speak additional utterances of the particular hotword; and

after obtaining the personalized hotword detection model that is trained to detect when the user speaks the particular hotword using the audio data corresponding to only the single utterance of the particular hotword by the user and without requiring the user to speak the additional utterances of the particular hotword, detecting, by the client device, the likely utterance of the particular hotword by the user in subsequently received audio data using the personalized hotword detection model.

3. The method of claim 2, wherein during an enrollment process, prompting the user to speak the particular hotword, and receiving the audio data corresponding to only the single utterance of the particular hotword by the user comprises:

prompting the user to speak one or more terms that trigger semantic interpretation of the one or more terms or one or more terms that follow the particular hotword.

4. The method of claim 2, wherein in response to receiving the audio data corresponding to only the single utterance of the particular hotword by the user, obtaining the personalized hotword detection model comprises:

generating the personalized hotword detection model after receiving the audio data corresponding to only the single utterance of the particular hotword by the user and without receiving additional audio data corresponding to another utterance of the particular hotword.

5. The method of claim 2, comprising:

ending the enrollment process after obtaining the personalized hotword detection model.

6. The method of claim 2, wherein the personalized hotword detection model is based at least on the single utterance and not based on another utterance of the particular hotword.

7. The method of claim 2, wherein using the personalized hotword detection model to detect the likely utterance of the particular hotword in the subsequently received audio data comprises:

receiving audio data corresponding to a subsequent utterance; and

determining whether the subsequent utterance likely includes the particular hotword based at least on the personalized hotword detection model.

8. The method of claim 7, comprising:

in response to determining whether the subsequent utterance likely includes the particular hotword based at least on the personalized hotword detection model, performing semantic interpretation on at least a portion of the subsequent utterance.

9. A system comprising:

one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising;

during an enrollment process, prompting, by a client device, a user to speak a particular hotword, and receiving, by the client device, audio data corresponding to only a single utterance of the particular hotword by the user;

in response to receiving the audio data corresponding to only the single utterance of the particular hotword by the user, obtaining, by the client device, a personalized hotword detection model, wherein the personalized hotword detection model is trained to detect a likely utterance of the particular hotword by the user using the audio data corresponding to only the single utterance of the particular hotword by the user and without requiring the user to speak additional utterances of the particular hotword; and

after obtaining the personalized hotword detection model that is trained to detect when the user speaks the particular hotword using the audio data corresponding to only the single utterance of the particular hotword by the user and without requiring the user to speak the additional utterances of the particular hotword, detecting, by the client device, the likely utterance of the particular hotword by the user in subsequently received audio data using the hotword detection model.

10. The system of claim 9, wherein during an enrollment process, prompting the user to speak the particular hotword, and receiving the audio data corresponding to only the single utterance of the particular hotword by the user comprises:

prompting the user to speak one or more terms that trigger semantic interpretation of the one or more terms or one or more terms that follow the particular hotword.

11. The system of claim 9, wherein in response to receiving the audio data corresponding to only the single utterance of the particular hotword by the user, obtaining the personalized hotword detection model comprises:

generating the personalized hotword detection model after receiving the audio data corresponding to only the single utterance of the particular hotword by the user and without receiving additional audio data corresponding to another utterance of the particular hotword.

12. The system of claim 9, the operations comprising:

ending the enrollment process after obtaining the personalized hotword detection model.

13. The system of claim 9, wherein the personalized hotword detection model is based at least on the single utterance and not based on another utterance of the particular hotword.

14. The system of claim 9, wherein using the personalized hotword detection model to detect the likely utterance of the particular hotword in the subsequently received audio data comprises:

receiving audio data corresponding to a subsequent utterance; and

determining whether the subsequent utterance likely includes the particular hotword based at least on the personalized hotword detection model.

15. The system of claim 14, comprising:

in response to determining whether the subsequent utterance likely includes the particular hotword based at least on the personalized hotword detection model, performing semantic interpretation on at least a portion of the subsequent utterance.

16. A non-transitory computer-readable medium storing software comprising instructions executable by one or more computers which, upon such execution, cause the one or more computers to perform operations comprising:

one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising;

during an enrollment process, prompting, by a client device, a user to speak a particular hotword, and receiving, by the client device, audio data corresponding to only a single utterance of the particular hotword by the user;

in response to receiving the audio data corresponding to only the single utterance of the particular hotword by the user, obtaining, by the client device, a personalized hotword detection model, wherein the personalized hotword detection model is trained to detect a likely utterance of the particular hotword by the user using the audio data corresponding to only the single utterance of the particular hotword by the user and without requiring the user to speak additional utterances of the particular hotword; and

after obtaining the personalized hotword detection model that is trained to detect when the user speaks the particular hotword using the audio data corresponding to only the single utterance of the particular hotword by the user and without requiring the user to speak the additional utterances of the particular hotword, detecting, by the client device, the likely utterance of the particular hotword by the user in subsequently received audio data using the hotword detection model.

17. The medium of claim 16, wherein during an enrollment process, prompting the user to speak the particular hotword, and receiving the audio data corresponding to only the single utterance of the particular hotword by the user comprises:

prompting the user to speak one or more terms that trigger semantic interpretation of the one or more terms or one or more terms that follow the particular hotword.

18. The medium of claim 16, wherein in response to receiving the audio data corresponding to only the single utterance of the particular hotword by the user, obtaining the personalized hotword detection model comprises:

generating the personalized hotword detection model after receiving the audio data corresponding to only the single utterance of the particular hotword by the user and without receiving additional audio data corresponding to another utterance of the particular hotword.

19. The medium of claim 16, the operations comprising:

ending the enrollment process after obtaining the personalized hotword detection model.

20. The medium of claim 16, wherein the personalized hotword detection model is based at least on the single utterance and not based on another utterance of the particular hotword.

21. The medium of claim 16, wherein using the personalized hotword detection model to detect a likely utterance of the particular hotword in subsequently received audio data comprises:

receiving audio data corresponding to a subsequent utterance; and

determining whether the subsequent utterance likely includes the particular hotword based at least on the personalized hotword detection model.