I found this from ~3 months ago, the recent addition to github: https://google.github.io/tacotron/publications/uncovering_latent_style_factors_for_expressive_s

I found this from ~3 months ago, the recent addition to github: https://google.github.io/tacotron/publications/uncovering_latent_style_factors_for_expressive_speech_synthesis/index.html https://google.github.io/tacotron/publications/uncovering_la... Seems to do with manipulating the speed styles in how we speak. Listen to the "neutral" blue lagoon, vs token 6 (tired sound), vs mix w/ token 7 (prominence)" (very confident). Paper: https://arxiv.org/pdf/1711.00520.pdf https://arxiv.org/pdf/1711.00520.pdf