Abstract: A method of building a new voice having a new timbre using a timbre vector space includes receiving timbre data filtered using a temporal receptive field. The timbre data is mapped in the timbre vector space. The timbre data is related to a plurality of different voices. Each of the plurality of different voices has respective timbre data in the timbre vector space. The method builds the new timbre using the timbre data of the plurality of different voices using a machine learning system.

Abstract: A method for online voice content moderation provides a multi-stage voice content analysis system. The system includes a pre-moderator stage having a toxicity scorer configured to provide a toxicity score for a given toxic speech content from a user. The toxicity score is a function of a platform content policy. The method generates a toxicity score for the given toxic speech content. The toxic speech content is provided to a moderator as a function of the toxicity score.

Abstract: A method watermarks speech data by using a generator to generate speech data including a watermark. The generator is trained to generate the speech data including the watermark. The training process generates first speech from the generator. The first speech data is configured to represent speech. The first speech data includes a candidate watermark. The training also produces an inconsistency message as a function of at least one difference between the first speech data and at least authentic speech data. The training further includes transforming the first speech data, including the candidate watermark, using a watermark robustness module to produce transformed speech data including a transformed candidate watermark. The transformed speech data includes a transformed candidate watermark. The training further produces a watermark-detectability message, using a watermark detection machine learning system, relating to one or more desirable watermark features of the transformed candidate watermark.

Abstract: A method of building a new voice having a new timbre using a timbre vector space includes receiving timbre data filtered using a temporal receptive field. The timbre data is mapped in the timbre vector space. The timbre data is related to a plurality of different voices. Each of the plurality of different voices has respective timbre data in the timbre vector space. The method builds the new timbre using the timbre data of the plurality of different voices using a machine learning system.

Abstract: A timbre vector space construction system for building a timbre vector space has an input. The input is configured to receive a first speech segment in a first voice and a second speech segment in a second voice. The system also includes a temporal receptive field to transform the first speech segment into a first plurality of analytical segments, and the second speech segment into a second plurality of analytical segments. Each of the first plurality of smaller analytical segments, and each of the second plurality of analytical segments have a frequency distribution that represents a different portion of the timbre data of the respective voices. The system also includes a machine learning system configured to map the first voice relative to the second voice in the timbre vector space as a function of the frequency distribution of the first plurality of analytical segments the second plurality of analytical segments.

Abstract: A method of building a new voice having a new timbre using a timbre vector space includes receiving timbre data filtered using a temporal receptive field. The timbre data is mapped in the timbre vector space. The timbre data is related to a plurality of different voices. Each of the plurality of different voices has respective timbre data in the timbre vector space. The method builds the new timbre using the timbre data of the plurality of different voices using a machine learning system.

Abstract: A method of building a speech conversion system uses target information from a target voice and source speech data. The method receives the source speech data and the target timbre data, which is within a timbre space. A generator produces first candidate data as a function of the source speech data and the target timbre data. A discriminator compares the first candidate data to the target timbre data with reference to timbre data of a plurality of different voices. The discriminator determines inconsistencies between the first candidate data and the target timbre data. The discriminator produces an inconsistency message containing information relating to the inconsistencies. The inconsistency message is fed back to the generator, and the generator produces a second candidate data. The target timbre data in the timbre space is refined using information produced by the generator and/or discriminator as a result of the feeding back.