CUSTOMIZING SPEECH-RECOGNITION DICTIONARIES IN A SMART-HOME ENVIRONMENT

• US 20160314782A1
• Filed: 04/21/2015
• Published: 10/27/2016
• Est. Priority Date: 04/21/2015
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Klimanis, Gints Valdis
• Granted Patent: US 10,079,012 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/07 to the speaker
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques

Abstract

A method for customizing speech-recognition dictionaries for different smart-home environments may include generating, at a smart-home device mounted in an enclosure, an acoustic impulse response for the enclosure. The method may also include receiving, by the smart-home device, an audio signal captured in the enclosure. The method may additionally include performing, by the smart-home device, a speech-recognition process on the audio signal using a second speech dictionary generated by convolving the acoustic impulse response with a first speech dictionary.

Claims

1. A method for customizing speech-recognition dictionaries for different smart-home environments, the method comprising:

generating, at a smart-home device mounted in an enclosure, an acoustic impulse response for the enclosure;

receiving, by the smart-home device, an audio signal captured in the enclosure; and

performing, by the smart-home device, a speech-recognition process on the audio signal using a second speech dictionary generated by convolving the acoustic impulse response with a first speech dictionary.

2. The method of claim 1, further comprising:

sending the acoustic impulse response from the smart-home device to a management server; and

receiving, by the smart-home device and from the management server, the second speech dictionary, wherein the management server convolved the acoustic impulse response captured by the smart-home device with the first speech dictionary.

3. The method of claim 1, further comprising:

generating, by the smart-home device, a sound approximating an ideal acoustic impulse; and

recording, by the smart-home device, an acoustic response of the enclosure to the sound approximating an ideal acoustic impulse.

4. The method of claim 3, wherein the acoustic impulse response for the enclosure is generated from the acoustic response of the enclosure.

5. The method of claim 3, wherein the sound approximating the ideal acoustic impulse comprises an electronic click or chirp generated by the smart-home device.

6. The method of claim 3, wherein generating the sound approximating the ideal acoustic impulse and recording the acoustic response of the enclosure occur during a learning interval of the smart-home device.

7. The method of claim 6, wherein the learning interval comprises a time interval within two weeks of an installation of the smart-home device.

8. A smart home device comprising:

a recording device;

one or more environmental sensors;

one or more processors;

one or more memory devices comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising;

generating an acoustic impulse response for the enclosure in which the smart-home device is installed;

receiving, through the recording device, an audio signal captured in the enclosure; and

performing, by the one or more processors, a speech-recognition process on the audio signal using a second speech dictionary generated by convolving the acoustic impulse response with a first speech dictionary.

9. The smart-home device of claim 8, further comprising a wireless network device, wherein the operations further comprise:

sending the acoustic impulse response from the smart-home device to a management server through the wireless network device; and

receiving, from the management server through the wireless network device, the second speech dictionary, wherein the management server convolves the acoustic impulse response captured by the smart-home device with the first speech dictionary to generate the second speech dictionary.

10. The smart-home device of claim 8, wherein the smart-home device is a hazard detector and the one or more environmental sensors comprises a smoke sensor.

11. The smart-home device of claim 8, wherein the smart-home device is a thermostat and the one or more environmental sensors comprises a temperature sensor.

12. The smart-home device of claim 8, wherein the enclosure comprises a room within a home.

13. The smart-home device of claim 8, wherein the operations further comprise:

receiving input from the one or more environmental sensors indicating that the enclosure is in a first state, wherein the second speech dictionary is used by the speech-recognition process when the enclosure is in the first state.

14. The smart-home device of claim 13, wherein the operations further comprise:

receiving input from the one or more environmental sensors indicating that the enclosure is in a second state;

generating a second acoustic impulse response for the enclosure in the second state;

receiving, through the recording device, a second audio signal captured in the enclosure while the enclosure is in the second state; and

performing, by the one or more processors, the speech-recognition process on the audio signal using a third speech dictionary generated by convolving the second acoustic impulse response with the first speech dictionary.

15. The smart-home device of claim 14, wherein:

when the enclosure is in the first state, a door or window of the enclosure is closed; and

when the enclosure is in the second state, the door or window of the enclosure is open.

16. A non-transitory, computer-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:

generating an acoustic impulse response for the enclosure in which the smart-home device is installed;

receiving, through a recording device, an audio signal captured in the enclosure; and

performing, by the one or more processors, a speech-recognition process on the audio signal using a second speech dictionary generated by convolving the acoustic impulse response with a first speech dictionary.

17. The non-transitory, computer-readable medium of claim 16, wherein the first speech dictionary is stored at the management server and available to a plurality of smart-home devices in a plurality of enclosures.

18. The non-transitory, computer-readable medium of claim 16, wherein the operations further comprise:

identifying a first user of the smart-home device, wherein the second speech dictionary is further personalized to the first user;

determining that the audio signal captured in the enclosure comprises human speech by the first user; and

using the second speech dictionary for the speech-recognition process in response to determining that the audio signal captured in the enclosure comprises human speech by the first user.

19. The non-transitory, computer-readable medium of claim 16, wherein the operations further comprise:

transmitting the acoustic impulse response to a second smart-home device in the enclosure over a local smart-home-device network.

20. The non-transitory, computer-readable medium of claim 16, wherein the operations further comprise:

transmitting the second speech dictionary to a second smart-home device in the enclosure over a local smart-home-device network.