Scaling and bias codes for modeling speaker-adaptive DNN-based speech synthesis systems

Speech samples for the paper "Scaling and bias codes for modeling speaker-adaptive DNN-based speech synthesis systems" which is presented at IEEE SLT 2018 - Workshop on Spoken Language Technology.

A pre-print version of this paper can be found at https://arxiv.org/abs/1807.11632

More publications (arxiv pre-print, posters,...) about speaker adaptation and speech synthesis can be found at my website.

WORLD vocoder

samples synthesized using WORLD vocoder

Wavenet vocoder

samples synthesized using Wavenet vocoder for selected strategies

1st sample

Non-linear Linear
10 320 10 320
Natural ► Play
Biasm ► Play ► Play
Bias ► Play ► Play not available not available
Scale not available not available not available not available
Affine not available not available ► Play ► Play
Level not available not available ► Play ► Play
Bottle not available not available ► Play ► Play

2nd sample

Non-linear Linear
10 320 10 320
Natural ► Play
Biasm ► Play ► Play
Bias ► Play ► Play not available not available
Scale not available not available not available not available
Affine not available not available ► Play ► Play
Level not available not available ► Play ► Play
Bottle not available not available ► Play ► Play

asdasdas