US 10614826

Patent Overview — US 10614826 B2
Source: https://patents.google.com/patent/US10614826/en

Title: System and method for voice-to-voice conversion.

Assignee/Applicant: Modulate Inc (original assignee Modulate Inc; assignment to Modulate, LLC recorded 2018-06-22; name change to MODULATE, INC. recorded 2020-01-21, per the listed reassignment entries).

Key dates: Priority date 2017-05-24; filing date 2018-05-24; publication of US20180342256A1 on 2018-11-29; granted 2020-04-07; adjusted expiration listed as 2038-10-14; legal status listed as Active. Inventors: William Carter Huffman and Michael Pappas.

Purpose and plain-language scope: The patent generally relates to voice conversion and, more particularly, to generating synthetic voice profiles. A method of building a speech conversion system receives source speech data representing a first speech segment of a source voice, and target timbre data relating to a target voice, the target timbre data being within a timbre space. A generative machine learning system produces first candidate speech data representing a first candidate speech segment in a first candidate voice, as a function of the source speech data and the target timbre data.

A discriminative machine learning system compares the first candidate speech data to the target timbre data with reference to timbre data of a plurality of different voices, determines at least one inconsistency, and produces an inconsistency message that is fed back to the generative machine learning system. The generative system then produces second candidate speech data representing a second candidate speech segment in a second candidate voice as a function of the inconsistency message. The target timbre data in the timbre space is refined using information produced by the generative and/or discriminative system as a result of the feeding back, and the source speech data is transformed into the target timbre.

The excerpt states the discriminative neural network produces the inconsistency message when there is less than a 95 percent confidence interval that the first candidate voice is the target voice, and that the second candidate speech segment provides a higher probability of being identified as the target voice by the discriminator than the first candidate speech segment.

What the excerpt does not establish: It provides only a truncated summary/definitions section — the full claims, detailed description, drawings, and figures are omitted, so claim scope cannot be verified here. It also does not establish litigation, validity, or prior art.