Change default voice for Kokoro TTS to af

#15
by hexgrad - opened

af is the new default voice for Kokoro TTS. It is the equally weighted mean of all other American female voices (except af_nicole, who is an ASMR exception):

  • af_0
  • af_alloy
  • af_bella
  • af_jessica
  • af_nova
  • af_river
  • af_sarah
  • af_sky

This helps address some of the "overfitting to one voice" concerns raised in closed PR #14, without the downsides of stochasticity. Also, af could simply be a better voice than its predecessor af_0.

TODO for @Pendrokar if this PR is accepted: purge and refresh cached samples. Let me know if you need my help with this.

Matching commit: https://huggingface.co/spaces/hexgrad/Kokoro-TTS/commit/1daab5c11c6981ae05062909f04ceebb23e0609b
[Edit] Removed "_gladiatrix" suffix: https://huggingface.co/spaces/hexgrad/Kokoro-TTS/commit/480b27e5343161f4552c8069867c727bff99f917

hexgrad changed pull request title from Change default voice for Kokoro TTS to af_gladiatrix to Change default voice for Kokoro TTS to af
Pendrokar changed pull request status to merged

Sign up or log in to comment