US 10,622,009 B1Granted

Methods for detecting double-talk

AssigneeAmazon Technologies, Inc.

InventorsXianxian Zhang, Philip Ryan Hilmes, Trausti Thor Kristjansson

Life of this patent

Estimating the term…

Loading patent…

Abstract

A system configured to improve double-talk detection. The system detects when double-talk is present in a voice conversation using two or more speaker models. The system extracts feature data from microphone audio data and compares the feature data to each speaker model. For example, the system may generate a first distance score indicating a likelihood that the feature data corresponds to a far-end speaker model and a second distance score indicating a likelihood that the feature data corresponds to a universal speaker model. The system may determine current system conditions based on the distance scores and may change settings to improve speech quality during the voice conversation. For example, during far-end single-talk the system may aggressively reduce an echo signal, whereas during near-end single-talk and double-talk the system may apply minimal echo cancellation to improve a quality of the local speech.

Claims19

Dependent claims

A computer-implemented method, the method comprising:

receiving, by a local device from a remote device, playback audio data;
sending the playback audio data to a loudspeaker of the local device to generate output audio;
determining microphone audio data using a microphone of the local device, the microphone audio data including a first representation of at least a portion of the output audio;
processing the microphone audio data using a first model to generate a first score indicating a first likelihood that far-end speech originating from the remote device is represented in a first portion of the microphone audio data, the first model configured to detect the far-end speech;
processing the microphone audio data using a second model to generate a second score indicating a second likelihood that first speech is represented in the first portion of the microphone audio data, the second model configured to detect the first speech;
determining that the first score is above a first threshold value, indicating that the far-end speech is represented in the first portion of the microphone audio data;
determining that the second score is below a second threshold value, indicating that near-end speech originating at the local device is not represented in the first portion of the microphone audio data;
determining that the first portion of the microphone audio data represents the far-end speech but not the near-end speech;
and
enabling, in response to determining that the first portion of the microphone audio data represents the far-end speech but not the near-end speech, residual echo suppression.

7 more elements

A computer-implemented method, the method comprising:

receiving first audio data associated with at least one microphone of a first device;
determining, using a first model, a first score indicating a first likelihood that the first audio data corresponds to far-end speech not originating at the first device;
determining, using a second model, a second score indicating a second likelihood that the first audio data corresponds to the second model;
determining that the first score satisfies a first condition, the first condition corresponding to the far-end speech being represented in the first audio data;
determining that the second score satisfies a second condition, the second condition corresponding to near-end speech not being represented in the first audio data, the near-end speech originating at the first device;
determining, based on the first score satisfying the first condition and the second score satisfying the second condition, that only the far-end speech is represented in the first audio data;
and
setting, in response to determining that only the far-end speech is represented in the first audio data, at least one first parameter.

5 more elements

A computer-implemented method, the method comprising:

receiving first audio data associated with at least one microphone of a first device;
determining, using a first model, a first score indicating a first likelihood that the first audio data corresponds to far-end speech not originating at the first device;
determining, using a second model, a second score indicating a second likelihood that the first audio data corresponds to the second model;
determining that the first score satisfies a first condition, the first condition corresponding to the far-end speech being represented in the first audio data;
determining that the second score satisfies a second condition, the second condition corresponding to near-end speech being represented in the first audio data, the near-end speech originating at the first device;
determining, based on the first score satisfying the first condition and the second score satisfying the second condition, that the near-end speech and the far-end speech are both represented in the first audio data;
and
setting, in response to determining that the near-end speech and the far-end speech are both represented in the first audio data, at least one first parameter.

5 more elements

Dependencies are inferred from the claim text.