|
|
--- |
|
|
license: cc-by-nc-4.0 |
|
|
language: |
|
|
- de |
|
|
base_model: |
|
|
- SWivid/F5-TTS |
|
|
pipeline_tag: text-to-speech |
|
|
--- |
|
|
|
|
|
# F5 TTS German Model |
|
|
|
|
|
This model was trained for 4.2 million steps on the german Mozilla common voice 19.0 recordings and an internal dataset. It is designed for text-to-speech synthesis in German. \ |
|
|
The command to train the model is: |
|
|
|
|
|
accelerate launch --mixed_precision=bf16 finetune_cli.py --exp_name F5TTS_Base --learning_rate 1.8e-05 --batch_size_per_gpu 8000 --batch_size_type frame --max_samples 0 --grad_accumulation_steps 1 --max_grad_norm 1 --epochs 40 --num_warmup_updates 2000 --save_per_updates 100000 --last_per_steps 10000 --dataset_name german_speak --finetune --pretrain ckpts/german_speak/model_last.pt --tokenizer pinyin --log_samples --logger wandb |
|
|
|
|
|
|
|
|
### Model Checkpoint |
|
|
The checkpoint supports German and can be [downloaded here](https://huggingface.co/hvoss-techfak/F5-TTS-German/blob/main/model_f5tts_german.pt). |
|
|
|
|
|
|
|
|
### Resources |
|
|
- [Tutorial video to train a new language model](https://www.youtube.com/watch?v=UO4usaOojys) |
|
|
|
|
|
### About Us |
|
|
Check out our website: [SCS Bielefeld University](https://scs.techfak.uni-bielefeld.de/) |
|
|
|
