Systems and methods for selective wake word detection using neural network models
Systems and methods for media playback via a media playback system include capturing sound data via a network microphone device and identifying a candidate wake word in the sound data. Based on identification of the candidate wake word in the sound data, the system selects a first wake-word engine from a plurality of wake-word engines. Via the first wake-word engine, the system analyzes the sound data to detect a confirmed wake word, and, in response to detecting the confirmed wake word, transmits a voice utterance of the sound data to one or more remote computing devices associated with a voice assistant service.
1. A method comprising:
capturing sound data via a network microphone device;
identifying, via the network microphone device, using a keyword spotting algorithm, a candidate wake word in the sound data, the network microphone device comprising a plurality of wake-world engines in a low-power or no-power state;
based on identification of the candidate wake word in the sound data, selecting a first wake-word engine from the plurality of wake-word engines, wherein the first wake-word engine is associated with a first voice assistant service and another of the plurality of wake-word engines is associated with a second voice assistant service different from the first, wherein selecting the first wake-word engine comprises activating the first wake-word engine from the low-power or no-power state to a high-power state, while the other(s) of the plurality of wake-word engines remain(s) in the low-power or no-power state;
with the first wake-word engine, analyzing the sound data to confirm detection of a wake word, wherein the first wake-word engine is configured to determine whether the candidate wake work is present in the sound data with a higher accuracy than the keyword spotting algorithm; and
in response to confirming the detection of the wake word, transmitting a voice utterance of the sound data to one or more remote computing devices associated with the first voice assistant service.
2. The method of claim 1 , wherein identifying the candidate wake word comprises determining a probability that the candidate wake word is present in the sound data.
3. The method of claim 1 , wherein the first wake-word engine is associated with the candidate wake word, and wherein another of the plurality of wake-word engines is associated with one or more additional wake words.
4. The method of claim 1 , wherein identifying the candidate wake word comprises applying a neural network model to the sound data.
5. The method of claim 4 , wherein the neural network model comprises a compressed neural network model stored locally on the network microphone device.
6. The method of claim 1 , further comprising, after transmitting the voice utterance, receiving, via the network microphone device, a selection of media content related to the voice utterance.
7. The method of claim 1 , wherein the plurality of wake-word engines comprises:
the first wake-word engine; and
a second wake-word engine configured to perform a local function of the network microphone device.
8. A network microphone device, comprising:
one or more processors;
at least one microphone; and
tangible, non-transitory, computer-readable media storing instructions executable by one or more processors to cause the network microphone device to perform operations comprising:
capturing sound data via the network microphone device;
identifying, via the network microphone device, using a keyword spotting algorithm, a candidate wake word in the sound data, the network microphone device comprising a plurality of wake-word engines in a low-power or no-power state;
based on identification of the candidate wake word in the sound data, selecting a first wake-word engine from the plurality of wake-word engines, wherein the first wake-word engine is associated with a first voice assistant service and another of the plurality of wake-word engines is associated with a second voice assistant service different from the first, and wherein selecting the first wake-word engine comprises activating the first wake-word engine from the low-power or no-power state to a high-power state while the other(s) of the plurality of wake-word engines remain(s) in the low-power or no-power state;
with the first wake-word engine, analyzing the sound data to confirm detection of a wake word, wherein the first wake-word engine is configured to determine whether the candidate wake word is present in the sound data with a higher accuracy than the keyword spotting algorithm; and
in response to confirming detection of the wake word, transmitting a voice utterance of the sound data to one or more remote computing devices associated with the first voice assistant service.
9. The network microphone device of claim 8 , wherein identifying the candidate wake word comprises determining a probability that the candidate wake word is present in the sound data.
10. The network microphone device of claim 8 , wherein the first wake-word engine is associated with the candidate wake word, and wherein another of the plurality of wake-word engines is associated with one or more additional wake words.
11. The network microphone device of claim 8 , wherein identifying the candidate wake word comprises applying a neural network model to the sound data.
12. The network microphone device of claim 11 , wherein the neural network model comprises a compressed neural network model stored locally on the network microphone device.
13. The network microphone device of claim 8 , wherein the operations further comprise, after transmitting the voice utterance, receiving, via the network microphone device, a selection of media content related to the voice utterance.
14. The network microphone device of claim 8 , wherein the plurality of wake-word engines comprises:
the first wake-word engine; and
a second wake-word engine configured to perform a local function of the network microphone device.
15. Tangible, non-transitory, computer-readable media storing instructions executable by one or more processors to cause a network microphone device to perform operations comprising:
capturing sound data via the network microphone device;
identifying, via the network microphone device, using a keyword spotting algorithm, a candidate wake word in the sound data, the network microphone device comprising a plurality of wake-word engines in a low-power or no-power state;
based on identification of the candidate wake word in the sound data, selecting a first wake-word engine from the plurality of wake-word engines, wherein the first wake-word engine is associated with a first voice assistant service and another of the plurality of wake-word engines is associated with a second voice assistant service different from the first, and wherein selecting the first wake-word engine comprises activating the first wake-word engine from the low-power or no-power state to a high-power state while the other(s) of the plurality of wake-word engines remain(s) in the low-power or no-power state;
with the first wake-word engine, analyzing the sound data to confirm detection of a wake word, wherein the first wake-word engine is configured to determine whether the candidate wake work is present in the sound data with a higher accuracy than the keyword spotting algorithm; and
in response to confirming detection of the wake word, transmitting a voice utterance of the sound data to one or more remote computing devices associated with the first voice assistant service.
16. The tangible, non-transitory, computer-readable media of claim 15 , wherein identifying the candidate wake word comprises determining a probability that the candidate wake word is present in the sound data.
17. The tangible, non-transitory, computer-readable media of claim 15 , wherein the first wake-word engine is associated with the candidate wake word, and wherein another of the plurality of wake-word engines is associated with one or more additional wake words.
18. The tangible, non-transitory, computer-readable media of claim 15 , wherein identifying the candidate wake word comprises applying a locally stored neural network model to the sound data.
19. The tangible, non-transitory, computer-readable media of claim 15 , wherein the operations further comprise, after transmitting the voice utterance, receiving, via the network microphone device, a selection of media content related to the voice utterance.
20. The tangible, non-transitory, computer-readable media of claim 15 , wherein the plurality of wake-word engines comprises:
the first wake-word engine; and
a second wake-word engine configured to perform a local function of the network microphone device.