US 2024/0404512 A1 · SYSTEM AND METHOD TO NATURALLY JOIN HUMAN TALKING SOUNDS

Abstract

The present disclosure describes systems and methods for naturally joining human talking sounds. An example method includes determining a first audio sequence and a second audio sequence from a full audio sequence based on a desired edit point. The method also includes determining a first separation point that defines a front portion of the first audio sequence and a back portion of the first audio sequence. The method yet further includes determining a second separation point that defines a front portion of the second audio sequence and a back portion of the second audio sequence, further defining an audio portion to be replaced. The method includes determining a regenerated audio portion, blending the regenerated audio portion with respective extensions of the front portion of the first audio sequence and the back portion of the second audio sequence, and replacing the audio portion to be replaced with the blended regenerated audio.

Claims20

Dependent claims

A method comprising:

determining a first audio sequence and a second audio sequence to join based on a desired edit point;
determining a first separation point that defines a front portion of the first audio sequence and a back portion of the first audio sequence;
determining a second separation point that defines a front portion of the second audio sequence and a back portion of the second audio sequence, wherein a portion of the full audio sequence between the first separation point and the second separation point defines an audio portion to be replaced, wherein determining the first separation point or determining the second separation point is based on identifying at least one silence interval;
determining a regenerated audio portion;
blending the regenerated audio portion with respective extensions of the front portion of the first audio sequence and the back portion of the second audio sequence to provide blended regenerated audio;
replacing the audio portion to be replaced with the blended regenerated audio.

3 more elements

A system comprising:

a sub-sequence separator configured to:
identify a first audio sequence and a second audio sequence from a full audio sequence defined by a desired edit point;
determine a first separation point that defines a front portion of the first audio sequence and a back portion of the first audio sequence;
and
determine a second separation point that defines a front portion of the second audio sequence and a back portion of the second audio sequence, wherein a portion of the full audio sequence between the first separation point and the second separation point defines an audio portion to be replaced;
a gap mask predictor configured to determine a gap duration of the audio portion to be replaced;
an audio representation recovery module configured to:
determine a regenerated audio portion;
and
blend the regenerated audio portion with respective extensions of the front portion of the first audio sequence and the back portion of the second audio sequence to provide blended regenerated audio;
and
an audio join module configured to:
replace the audio portion to be replaced with the blended regenerated audio.

10 more elements

Dependencies are inferred from the claim text.

System and method to naturally join human talking sounds

Life of this patent

System and method to naturally join human talking sounds

Life of this patent

Abstract

Claims20