System for processing voice requests including wake word verification

Patent metadata

Field	Value
Publication number	US-12488787-B2
Application number	US-202218090064-A
Country	US
Kind code	B2
Filing date	Dec 28, 2022
Priority date	Dec 28, 2022
Publication date	Dec 2, 2025
Grant date	Dec 2, 2025

Abstract

A system for processing voice requests includes a voice assistant manager and a plurality of voice assistants. The voice assistant manager detects a wake word in an utterance and communicates the utterance to a voice assistant of the plurality of voice assistants. In some embodiments, the voice assistant may verify the detected wake word and communicate with a cloud service, which may also verify the detected wake word and generate a response to the utterance. In some embodiments, the voice assistant manager may activate or deactivate one or more of the voice assistants.

First claim

What is claimed is:

1 . A system for processing voice requests, the system comprising:
a voice assistant manager; a plurality of voice assistants; a processor; and memory coupled to the processor, wherein the memory stores instructions that, when executed by the processor, cause the voice assistant manager to:
receive an utterance from a user,
detect a wake word in the utterance using a first wake word detection model,
based on the wake word, identify a called assistant from the plurality of voice assistants, and
communicate the utterance to the called assistant,
wherein the instructions, when executed by the processor, cause the called assistant to:
receive the utterance from the voice assistant manager; and verify the wake word in the received utterance using a second wake word detection model,
wherein the second wake word detection model is trained to recognize one or more wake words,
wherein verifying the wake word comprises inputting the utterance into the second wake word detection model, and
wherein the system is further configured to generate a response to the utterance and transmit the response to the user.

2 . The system of claim 1 , wherein the instructions, when executed by the processor, further cause the voice assistant manager to:
deactivate the called assistant; and listen for a second utterance.

3 . The system of claim 1 , wherein the instructions, when executed by the processor, further cause the voice assistant manager to:
determine whether the called assistant is active; and in response to determining that the called assistant is not active, activate the called assistant before communicating the utterance to the called assistant.

4 . The system of claim 3 , wherein the instructions, when executed by the processor, further cause the voice assistant manager to, before activating the called assistant, deactivate a second assistant of the plurality of voice assistants.

5 . The system of claim 1 , wherein the called assistant is communicatively coupled to the voice assistant manager via a Matter network.

6 . The system of claim 1 , further comprising a cloud service associated with the called assistant;
wherein the instructions, when executed by the processor, further cause the called assistant to, in response to successfully verifying the wake word, communicate the utterance to the cloud service; and wherein the cloud service verifies the wake word.

7 . The system of claim 6 , wherein the cloud service, in response to successfully verifying the wake word, processes a request of the utterance.

8 . The system of claim 6 , wherein the cloud service, in response to failing to verify the wake word, returns an error to the called assistant and deletes data associated with the utterance.

9 . The system of claim 6 , wherein the instructions, when executed by the processor, cause the called assistant to:
receive the response from the cloud service; and communicate the response to one or more of the user or the voice assistant manager.

10 . The system of claim 6 , wherein one or more of communicating the utterance to the cloud service or communicating the utterance to the called assistant comprises sending a plurality of audio files, the plurality of audio files including an encrypted audio file and an unencrypted audio file;
wherein the unencrypted audio file includes the wake word; and wherein the encrypted audio file includes one or more of the utterance or a request of the utterance.

11 . The system of claim 1 , wherein the instructions, when executed by the processor, further cause the voice assistant manager to:
receive a subscription request from the called assistant, the subscription request including the wake word; associate the wake word with the called assistant; and train a machine learning model to recognize the wake word.

12 . The system of claim 1 , further comprising a computing device, wherein the computing device includes the voice assistant manager, the plurality of voice assistants, the processor, and the memory.

13 . The system of claim 12 , wherein the computing device includes a screen displaying a user interface; and
wherein the user interface includes a plurality of voice assistant icons, wherein each icon of the plurality of voice assistant icons corresponds with a voice assistant of the plurality of voice assistants.

14 . A method for processing voice requests, the method comprising:
receiving an utterance from a user; detecting, at a voice assistant manager, a wake word in the utterance using a first wake word detection model; identifying, from a plurality of voice assistants, a called assistant associated with the wake word; communicating the utterance to the called assistant; detecting, at the called assistant, the wake word in the utterance using a second wake word detection model, wherein the second wake word detection model is trained to recognize one or more wake words, and wherein detecting the wake word at the called assistant comprises inputting the utterance into the second wake word detection model; generating a response to the utterance; and transmitting the response to the user.

15 . The method of claim 14 , further comprising:
transmitting the utterance to a cloud service; and detecting, at the cloud service, the wake word in the utterance; wherein generating the response to the utterance is performed at the cloud service.

16 . The method of claim 14 , further comprising:
prior to communicating the utterance to the called assistant, activating the called assistant; determining that the called assistant finished processing the utterance; and deactivating the called assistant.

17 . The method of claim 14 , further comprising:
subscribing the called assistant; wherein subscribing the called assistant comprises (i) receiving one or more wake words associated with the called assistant, the one or more wake words associated with the called assistant including the wake word and (ii) training a machine learning model to detect the one or more wake words, wherein detecting, at the voice assistant manager, the wake word in the utterance comprises inputting the utterance into the machine learning model.

18 . A device for processing voice commands, the device comprising:
a processor; and memory coupled to the processor, the memory storing instructions that, when executed by the processor cause the device to:
receive an utterance;
detect a wake word in the utterance using a first wake word detection model;
identify, from a plurality of voice assistants, a called assistant associated with the wake word;
communicate the utterance to the called assistant;

detecting, at the called assistant, the wake word in the utterance using a second wake word detection model, wherein the second wake word detection model is trained to recognize one or more wake words, and wherein detecting the wake word at the called assistant comprises inputting the utterance into the second wake word detection model;
generate, at the called assistant, a response to the utterance; and
transmit the response to a user.

19 . The device of claim 18 , wherein detecting the wake word in the utterance using the first wake word detection model is performed at a voice assistant manager; and
wherein the instructions, when executed by the processor, further cause the device to: prior to communicating the utterance to the called assistant, activate the called assistant; and after generating the response

Assignees

• Spotify Ab

Inventors

• Bromand Daniel
• Roth Björn Erik
• Priem Nick

Classifications

• G10L15/22

  Procedures used during a speech recognition process, e.g. man-machine dialogue · CPC title

• G10L2015/223

  Execution procedure of a spoken command · CPC title

• G10L2015/088

  Word spotting · CPC title

• G10L15/08 Primary

  Speech classification or search · CPC title