Hotword detection on multiple devices
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for hotword detection on multiple devices are disclosed. In one aspect, a method includes the actions of receiving, by a first computing device, audio data that corresponds to an utterance. The actions further include determining a first value corresponding to a likelihood that the utterance includes a hotword. The actions further include receiving a second value corresponding to a likelihood that the utterance includes the hotword, the second value being determined by a second computing device. The actions further include comparing the first value and the second value. The actions further include based on comparing the first value to the second value, initiating speech recognition processing on the audio data.
1. A computer-implemented method comprising:
receiving, by a computing device that is in a low power mode and that is configured to exit a low power mode upon detecting an utterance of a particular, predefined hotword using an on-device hotword detector, audio data that corresponds to an utterance of the particular, predefined hotword;
while the computing device remains in the low power mode, and in response to receiving the audio data that corresponds to the utterance of the particular, predefined hotword, transmitting, by the computing device and to another computing device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, an output of processing the audio data using the on-device hotword detector;
while the computing device remains in low power mode, receiving, by the computing device and from the other computing device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, an additional output of processing the audio data; and
after transmitting the output of processing the audio data using the on-device hotword detector and after receiving the additional output of processing the audio data from the other using device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, determining, by the computing device, to remain in the low power mode.
2. The method of claim 1 ,
wherein determining to remain in the low power mode is based at least in part on receiving the additional output of processing the audio from the other using device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword.
3. The method of claim 1 , comprising:
determining a hotword confidence score that is based on the audio data that corresponds to the utterance of the particular, predefined hotword and that reflects a likelihood that the audio data includes the particular, predefined hotword,
wherein the output of processing the audio data using the on-device hotword detector includes the hotword confidence score.
4. The method of claim 1 , comprising:
determining a hotword confidence score that is based on the audio data that corresponds to the utterance of the particular, predefined hotword and that reflects a likelihood that the audio data includes the particular, predefined hotword; and
determining that the hotword confidence score satisfies a threshold,
wherein transmitting the output of processing the audio data using the on-device hotword detector is based on determining that the hotword confidence score satisfies the threshold.
5. The method of claim 1 , wherein the computing device transmits the output of processing the audio data using the on-device hotword detector without performing speech recognition on the audio data that corresponds to an utterance of the particular, predefined hotword.
6. The method of claim 1 , wherein:
the computing device transmits the output of processing the audio data using the on-device hotword detector for a particular amount of time, and
the computing device determines to remain in the low power mode after transmitting the output of processing the audio data using the on-device hotword detector for the particular amount of time.
7. The method of claim 1 , comprising:
determining a hotword confidence score that is based on the audio data that corresponds to an utterance of the particular, predefined hotword and that reflects a likelihood that the audio data includes the particular, predefined hotword,
wherein receiving, by the computing device and from the other computing device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, an additional output of processing the audio data comprises receiving, by the computing device and from the other computing device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, a hotword confidence score that is based on the audio data that corresponds to the utterance of the particular, predefined hotword and that reflects an additional likelihood that the audio data includes the particular, predefined hotword,
wherein the method comprises comparing the hotword confidence score with the additional hotword confidence score, and
wherein determining to remain in the low power mode is based on comparing the hotword confidence score with the additional hotword confidence score.
8. The method of claim 1 , wherein:
the other computing device is in a vicinity of the computing device,
the computing device receives the utterance of the particular, predefined hotword through a microphone of the computing device,
the other computing device receives the utterance of the particular, predefined hotword through another microphone of the computing device, and
the additional output of processing the audio data is based on processing, by the other computing device, the utterance of the particular, predefined hotword received by the other computing device.
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:
receiving, by a computing device that is in a low power mode and that is configured to exit a low power mode upon detecting an utterance of a particular, predefined hotword using an on-device hotword detector, audio data that corresponds to an utterance of a particular, predefined hotword;
while the computing device remains in the low power mode, and in response to receiving the audio data that corresponds to the utterance of the particular, predefined hotword, transmitting, by the computing device and to another computing device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, an output of processing the audio data using the on-device hotword detector;
while the computing device remains in low power mode, receiving, by the computing device and from the other computing device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, an additional output of processing the audio data; and
after transmitting the output of processing the audio data using the on-device hotword detector and after receiving the additional output of processing the audio data from the other using device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, determining, by the computing device, to remain in the low power mode.
10. The system of claim 9 ,
wherein determining to remain in the low power mode is based at least in part on receiving the additional output of processing the audio data from the other using device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword.
11. The system of claim 9 , wherein the operations further comprise:
determining a hotword confidence score that is based on the audio data that corresponds to the utterance of the particular, predefined hotword and that reflects a likelihood that the audio data includes the particular, predefined hotword,
wherein the output of processing the audio data using the on-device hotword detector includes the hotword confidence score.
12. The system of claim 9 , wherein the operations further comprise:
determining a hotword confidence score that is based on the audio data that corresponds to the utterance of the particular, predefined hotword and that reflects a likelihood that the audio data includes the particular, predefined hotword; and
determining that the hotword confidence score satisfies a threshold,
wherein transmitting the output of processing the audio data using the on-device hotword detector is based on determining that the hotword confidence score satisfies the threshold.
13. The system of claim 9 , wherein the computing device transmits the output of processing the audio data using the on-device hotword detector without performing speech recognition on the audio data that corresponds to an utterance of the particular, predefined hotword.
14. The system of claim 9 , wherein:
the computing device transmits the output of processing the audio data using the on-device hotword detector for a particular amount of time, and
the computing device determines to remain in the low power mode after transmitting the output of processing the audio data using the on-device hotword detector for the particular amount of time.
15. The system of claim 9 , wherein the operations further comprise:
determining a hotword confidence score that is based on the audio data that corresponds to an utterance of the particular, predefined hotword and that reflects a likelihood that the audio data includes the particular, predefined hotword,
wherein receiving, by the computing device and from the other computing device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, an additional output of processing the audio data comprises receiving, by the computing device and from the other computing device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, a hotword confidence score that is based on the audio data that corresponds to the utterance of the particular, predefined hotword and that reflects an additional likelihood that the audio data includes the particular, predefined hotword,
wherein the operations further comprise comparing the hotword confidence score with the additional hotword confidence score, and
wherein determining to remain in the low power mode is based on comparing the hotword confidence score with the additional hotword confidence score.
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:
receiving, by a computing device that is in a low power mode and that is configured to exit a low power mode upon detecting an utterance of a particular, predefined hotword using an on-device hotword detector, audio data that corresponds to an utterance of a particular, predefined hotword;
while the computing device remains in the low power mode, and in response to receiving the audio data that corresponds to the utterance of the particular, predefined hotword, transmitting, by the computing device and to another computing device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, an output of processing the audio data using the on-device hotword detector;
while the computing device remains in low power mode, receiving, by the computing device and from the other computing device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, an additional output of processing the audio data; and
after transmitting the output of processing the audio data using the on-device hotword detector and after receiving the additional output of processing the audio data from the other using device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword, determining, by the computing device, to remain in the low power mode.
17. The medium of claim 16 ,
wherein determining to remain in the low power mode is based at least in part on receiving the additional output of processing the audio from the other using device that is configured to exit a low power mode upon detecting an utterance of the particular, predefined hotword.
18. The medium of claim 16 , wherein the operations further comprise:
determining a hotword confidence score that is based on the audio data that corresponds to the utterance of the particular, predefined hotword and that reflects a likelihood that the audio data includes the particular, predefined hotword,
wherein the output of processing the audio data using the on-device hotword detector includes the hotword confidence score.
19. The medium of claim 16 , wherein the operations further comprise:
determining a hotword confidence score that is based on the audio data that corresponds to the utterance of the particular, predefined hotword and that reflects a likelihood that the audio data includes the particular, predefined hotword; and
determining that the hotword confidence score satisfies a threshold,
wherein transmitting the output of processing the audio data using the on-device hotword detector is based on determining that the hotword confidence score satisfies the threshold.
20. The medium of claim 16 , wherein the computing device transmits the output of processing the audio data using the on-device hotword detector without performing speech recognition on the audio data that corresponds to an utterance of the particular, predefined hotword.
21. The medium of claim 16 , wherein:
the computing device transmits the output of processing the audio data using the on-device hotword detector for a particular amount of time, and
the computing device determines to remain in the low power mode after transmitting the output of processing the audio data using the on-device hotword detector for the particular amount of time.