F5-TTS Finetune Web UI The E2-F5-TTS project has a huge update. Now anyone can easily finetune the model with their own audio using a Gradio app! It automatically transcribes the audio clips using whipser, and seems to work for most languages. Here's a Korean example.

This new Finetuning UI is now included in the official Github repository, so all you need to do is pull in the updates in Pinokio (or install fresh)

Also, you can even use @weights_biases for logging while training, simply by setting the WANDB_API_KEY environment variable in the configure tab. If you don't see this field after updating, you can edit the file manually to add the entry, or just reinstall from scratch.

Also, I'm proud to say this feature was my contribution :) By providing an environment variable based configuration, Now it's possible to deploy the training UI to the cloud with full wandb support as well as even packaged as a docker container if you want. cc: @altryne

I know many are waiting for multi-language support, and this UI will make it easy to train. I've heard people training even on their PC (not enterprise machines). To learn more, join this Github discussion and ask questions or share your results.

In the video i only showed uploading one audio clip but you can actually upload multiple (also my contribution haha) When you do so, it will automatically chunk up the audio clips into small .wav file pieces internally (as seen in the video as well)

Also @jpgallegoar probably knows a lot about this since he's been training a spanish version, so ask him for advice too if you're trying to train a custom language model (and of course read the full thread here as well).

Oh just to clarify before people misunderstand, the fine tune UI was not my work. Tons of other talented devs contributing to the project. I just pitched in with adding a proper wandb support so the training app can be deployed anywhere easily with just environment variable 😅

Note that there are two separate gradio apps. One is the original UI. And the other is for training. You need to launch the training UI

Many people asking what languages are supported. The transcription is automatically carried out using Whisper-Large-V3-Turbo, which supports 99 languages according to the huggingface page.

see