Adding custom models to TTS engines (F5 TTS)

[M4x-P0wer]

Hello alltalk Community,

I am wondering if it is possible to add a custom model to the available TTS engines.
As an example, I have a HF F5-TTS model (https://huggingface.co/marduk-ra/F5-TTS-German) that I would like to use via alltalk.

Is this possible?
I am unable to select the model if I try to add the related files to the directory.

[erew123]

Hi @M4x-P0wer

Yes you should be able to add any model to any of the TTS engines. I go by what the developer of the TTS engine says you need for file names etc:

https://github.com/erew123/alltalk_tts/blob/alltalkbeta/system/tts_engines/f5tts/model_engine.py#L402-L410

So I dont know which one of those models you are wanting to use, but you would need to rename them from f5_tts_german_1010000.safetensors to model_f5_tts_german_1010000.safetensors (basically add the word model to the start as AllTalk is looking for a model file with the word model at the start)

Make sure you also have the vocab.txt in the same folder.

Finally you will need the vocos folder too with its 2x files in. Im assuming that using the default ones from the default model will be fine.

[M4x-P0wer]

Thank you @erew123 and a happy new year!
Renaming the model (model_*) and directory (tts_v *) made it work.

[M4x-P0wer]

Renaming the model (model_) and directory (_tts_v *) made it work.