Detecting self-generated wake expressions
A speech-based audio device may be configured to detect a user-uttered wake expression. For example, the audio device may generate a parameter indicating whether output audio is currently being produced by an audio speaker, whether the output audio contains speech, whether the output audio contains a predefined expression, loudness of the output audio, loudness of input audio, and/or an echo characteristic. Based on the parameter, the audio device may determine whether an occurrence of the predefined expression in the input audio is a result of an utterance of the predefined expression by a user.
1. A system comprising:
a housing;
two or more microphones proximate a top of the housing;
one or more audio speakers disposed proximate a bottom of the housing and directed at least partly away from the one or more microphones;
one or more processors; and
non-transitory computer-readable media storing instructions that, when executed by the one or more processors, cause the system to:
generate, using the two or more microphones, a plurality of audio signals each corresponding to a sound;
determine one or more parameters associated with the plurality of audio signals;
determine, based at least in part on inputting the one or more parameters into a machine learned model trained to output an indication of trigger expressions within audio signals, a number of audio signals of the plurality of audio signals that include a trigger expression; and
determine, based at least in part on the number of audio signals, that the sound originated from the one or more audio speakers.
2. The system of claim 1 , wherein a first parameter of the one or more parameters corresponds to an audio input characteristic and a second parameter of the one or more parameters corresponds to a device operation characteristic.
3. The system of claim 2 wherein the audio input characteristic comprises an echo characteristic associated with the plurality of audio signals or a loudness characteristic associated with the plurality of audio signals.
4. The system of claim 2 , wherein the device operation characteristic comprises a presence of the one or more audio speakers, a loudness characteristic of sound generated by the one or more audio speaker, or an amount of echo reduction performed by the one or more processors.
5. The system of claim 1 , wherein the instructions, when executed by the one or more processors, further cause the system to:
determine a pattern of input signals based at least partly on the one or more parameters; and
generate a reference file based at least partly on the pattern of input signals.
6. The system of claim 5 , wherein the instructions, when executed by the one or more processors, further cause the system to:
analyze, using the one or more parameters, the plurality of audio signals to generate text data corresponding to the plurality of audio signals; and
cause, using the one or more audio speakers and based at least partly on the text data, output of second audio data;
generate, using the two or more microphones, third audio data; and
determine a second confidence value indicating a comparison of the third audio data to the reference file, the second audio data being based at least partly on the second confidence value.
7. The system of claim 1 wherein:
a first parameter of the one or more parameters indicates a low audio speaker output loudness;
a second parameter of the one or more parameters indicates a high input audio loudness; and
the instructions, when executed by the one or more processors, further cause the system to determine whether to analyze content of the plurality of audio signals based at least partly on the first parameter and the second parameter.
8. The system of claim 1 , wherein:
a first parameter of the one or more parameters indicates a low degree of echo cancellation;
a second parameter of the one or more parameters indicates that the one or more audio speakers are producing sound; and
the instructions, when executed by the one or more processors, further cause the system to determine whether to analyze content of the first audio data plurality of audio signals based at least partly on the first parameter and the second parameter.
9. A device comprising:
a housing;
two or more microphones disposed proximate a top of the housing;
one or more audio speakers disposed proximate a bottom of the housing and directed at least partly away from the one or more microphones;
one or more processors; and
non-transitory computer-readable media storing instructions that, when executed by the one or more processors, cause the device to:
generate, using the two or more microphones, a plurality of audio signals each corresponding to a sound;
determine one or more parameters associated with the plurality of audio signals;
determine, based at least in part on inputting the one or more parameters into a machine learned model trained to output an indication of trigger expressions within audio signals, a number of audio signals of the plurality of audio signals that a include a trigger expression; and
determine, based at least in part on the number of audio signals, that the sound originated from the one or more audio speakers.
10. The device of claim 9 , wherein the instructions, when executed by the one or more processors, further cause the device to:
detect an echo associated with the plurality of audio signals;
detect a speaker loudness associated with the first audio data plurality of audio signals; or determine operational information of the device associated with the plurality of audio signals.
11. The device of claim 9 , wherein the instructions, when executed by the one or more processors, further cause the device to:
determine a first parameter based at least partly on content of the plurality of audio signals;
determine a second parameter based at least partly on an echo associated with generating the plurality of audio signals;
determine a third parameter based at least partly on a loudness associated with the plurality of audio signals;
determine a fourth parameter based at least partly on operational information of the device; and
generate a reference parameter based at least partly on the first parameter, the second parameter, the third parameter, or the fourth parameter.
12. The device of claim 11 , wherein the instructions, when executed by the one or more processors, further cause the device to:
generate, using the one or more audio speakers and as part of a device initialization, a reference sound; and
detect, using the two or more microphones, the reference sound, wherein the reference parameter is based at least partly on the reference sound.
13. A method implemented at least in part by a device, the method comprising:
determining a plurality of parameters associated with plurality of audio signals each corresponding to a sound detected by two or more microphones disposed proximate a top of a housing of the device;
determine, based at least in part on inputting the plurality of parameters into a machine learned model trained to output an indication of trigger expressions within audio signals, a number of audio signals of the plurality of audio signals that include a trigger expression; and
determine, based at least in part on the number of audio signals, that the sound originated from one or more speakers.
14. The method of claim 13 , wherein the one or more speakers output second audio comprises content from a third-party application.
15. The method of claim 13 , wherein the device comprises a speech interface device in communication with a server device.
16. The method of claim 13 , wherein the plurality of parameters are based at least partly on an amount of echo associated with the plurality of audio signals.
17. The method of claim 13 , wherein the plurality of parameters are based at least partly on a loudness associated with an output at the speaker.
18. The method of claim 13 , wherein the plurality of parameters are based at least partly on whether the speaker or a text-to-speech algorithm generates an output.