coqui-ai / TTS

🐸💬 - a deep learning toolkit for Text-to-Speech, battle-tested in research and production
http://coqui.ai
Mozilla Public License 2.0
34.99k stars 4.26k forks source link

Can't train a model, "[!] {name} is not a valid value" #413

Closed smaybius closed 3 years ago

smaybius commented 3 years ago

I'm trying to train a model with my own dataset, and I got this error. The same thing applied when I used the default LJSpeech dataset: https://pastebin.com/WugD8rZt

lexkoro commented 3 years ago

can you share your config?

smaybius commented 3 years ago

can you share your config?

https://pastebin.com/3hZqLcaa

lexkoro commented 3 years ago

"model": "ljspeech_test" is the problem. It has to be a valid tts model -> 'tacotron', 'tacotron2', 'glow_tts', 'speedy_speech'.

smaybius commented 3 years ago

"model": "ljspeech_test" is the problem. It has to be a valid tts model -> 'tacotron', 'tacotron2', 'glow_tts', 'speedy_speech'.

It fixed it, but RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 4.00 GiB total capacity; 2.24 GiB already allocated; 11.70 MiB free; 2.35 GiB reserved in total by PyTorch) This is on a laptop GTX 1650 and I don't want to invest in an eGPU just for this

erogol commented 3 years ago

you should reduce the batch_size tha. Try smaller values.