Wakeword detection
AssigneeAmazon Technologies, Inc.
InventorDeepak Yavagal
Life of this patent
Estimating the term…
Abstract
A device having a voice-based interface activates or “wakes” when it detects an utterance that includes a wakeword. The device includes a digital-signal processor (DSP) that implements a first, lower-power speech-processing component that detects when captured audio includes the wakeword; the first speech-processing component then activates a second, higher-accuracy wakeword detection component that confirms that the captured audio includes the wakeword. The device may configure a wakeword-detection parameter, speaker-identification parameter, and/or wakeword-detection model of the first speech-processing component based on device-use data corresponding to the device, such as an environment of the device and/or a usage pattern of the user of the device.
Claims25
A method for detecting a wakeword using a voice-controlled device, the method comprising:
- receiving device-use data corresponding to a location of the voice-controlled device;
- determining that the location is a public location;
- based at least in part on determining that the location is the public location, sending, to a first speech-processing component implemented at least in part by a digital signal processor (DSP) of the voice-controlled device, a command from a second speech-processing component to lower a wakeword-detection threshold of the first speech-processing component;
- receiving, from a microphone, audio data representing a user input;
- determining, using the first speech-processing component, that the audio data includes a representation of a wakeword;
- based at least in part on determining that the audio data includes the representation of the wakeword, sending the audio data to at least one central processing unit (CPU) of the voice-controlled device, the at least one CPU different than the DSP;
- determining, using the second speech-processing component, that the audio data includes the representation of the wakeword, wherein the second speech-processing component is implemented at least in part by the at least one CPU;
- sending, to at least one component of a speech processing system that is located remotely from the voice-controlled device, the audio data;
- receiving, from the speech processing system, content data responsive to the user input;
- and
- outputting content corresponding to the content data.
A method, comprising:
- determining device-use data corresponding to a voice-controlled device;
- determining a wakeword-detection parameter based at least in part on the device-use data, the wakeword-detection parameter corresponding to a wakeword-detection sensitivity;
- based at least in part on an indication received by a first speech-processing component of the voice-controlled device from a second speech-processing component of the voice-controlled device, determining an adjusted wakeword-detection parameter;
- receiving audio data corresponding to an utterance;
- determining presence of a wakeword represented by the audio data using the first speech-processing component, the audio data, and the adjusted wakeword-detection parameter;
- and
- determining, using a second speech-processing component, that the audio data includes a representation of the wakeword, wherein the second speech-processing component is different from the first speech-processing component.
A system comprising:
- at least one processor;
- and
- at least one memory including instructions that, when executed by the at least one processor, cause the system to:
- determine device-use data corresponding to a voice-controlled device;
- determine a speaker-identification parameter based at least in part on the device-use data, the speaker-identification parameter corresponding to a speaker-identification sensitivity;
- causing a first speech-processing component of the voice-controlled device to operate using the speaker-identification parameter;
- receive audio data corresponding to an utterance;
- determine, using the first speech-processing component and the speaker-identification parameter, that the audio data corresponds to user data;
- determine, using a second speech-processing component of the voice-controlled device, that the audio data corresponds to the user data, wherein the second speech-processing component is different from the first speech-processing component;
- and
- based at least in part on using the first speech-processing component and the second speech-processing component to determine that the audio data corresponds to the user data, determining that the utterance originated from a particular user.
Dependencies are inferred from the claim text.