docs SpeechT5 tts, CMU Arctic speaker embeddings

Hello, I can't get the speecht5 tts option to work properly. No matter what embedding name (bdl, slt,...) I set there is no change. Only a young woman's voice with much grain, noice in the background. Am I using wrong embedding names ? Is there something I am not paying attention to ? Have a great day.

Dec 18 '24 22:12 User110112

Ensure the voice name has been set in two places:

Settings -> Audio tab: 2: Admin Panel -> Settings -> Audio tab:

I will see about updating the documentation regarding this if this helps you!

Dec 18 '24 22:12 silentoplayz

@silentoplayz thank your for the answer. For some reason the voice does not change, no matter what speaker name I use. It's always the woman's voice. Do you have further recommendations ?

Dec 19 '24 13:12 User110112

Can't reproduce on 0.5.3 on MacOS running pipx installed though. Seems to work fine with OpenAI

Jan 03 '25 08:01 richtong

I'm having the same issue, and I think @User110112 is referring to the voices available when using the local (transformers) TTS

Feb 08 '25 21:02 orrinwitt

Specifying the full name of the dataset as TTS Model worked for me, e.g. cmu_us_clb_arctic-wav-arctic_a0001.

May 26 '25 05:05 ItsJustRuby