README.md

# Bangla VITS TTS Fine-tuning

Fine-tuning Coqui TTS `tts_models/bn/custom/vits-male` on custom Bangla voice dataset.


## Setup

### 1. Create Environment
```bash
conda create -n tts-bn python=3.10 -y
conda activate tts-bn
```

### 2. Install PyTorch (do this first)
```bash
pip install torch==2.12.0 torchaudio==2.11.0 --index-url https://download.pytorch.org/whl/cu118
or
pip install torch==2.7.1+cu118 torchaudio==2.7.1+cu118 --index-url https://download.pytorch.org/whl/cu118
```

### 3. Install Dependencies
```bash
pip install -r requirements.txt
```


## Training

### Start Fine-tuning
```bash
CUDA_VISIBLE_DEVICES=0 python -m TTS.bin.train_tts \
    --config_path configs/bangla.json \
    --restore_path your_file_location/tts/tts_models--bn--custom--vits-male/model_file.pth
```

### Continue Training from Checkpoint
```bash
CUDA_VISIBLE_DEVICES=0 python -m TTS.bin.train_tts \
    --config_path configs/bangla.json \
    --continue_path outputs/YOUR_RUN_FOLDER/
```

### inject your cleaner into the TTS package
```bash
CLEANERS_PATH=$(python -c "import TTS.tts.utils.text.cleaners as c; import inspect; print(inspect.getfile(c))")

cat bangla_cleaners.py >> $CLEANERS_PATH

# Verify it worked
python -c "from TTS.tts.utils.text import cleaners; print(hasattr(cleaners, 'bangla_cleaners'))"
```

### Monitor Training
```bash
tensorboard --logdir=outputs/ --port=8080
```

## Inference
```bash
python inference.py
```

### Serve Output Files
```bash
python -m http.server 8080
```
### Tunneling with ngrok
```bash
ngrok http --domain=hawkeyes.ngrok.app 8080
```