# Luganda TTS (lug-tts-v4)

Luganda Text-to-Speech model trained with NVIDIA NeMo.

## Models
- `luganda_fastpitch_final.nemo` - FastPitch spectrogram generator  
- `luganda_hifigan_final.nemo` - HiFi-GAN vocoder

## Usage
```python
from nemo.collections.tts.models import FastPitchModel, HifiGanModel
import soundfile as sf

fastpitch = FastPitchModel.restore_from("luganda_fastpitch_final.nemo")
hifigan = HifiGanModel.restore_from("luganda_hifigan_final.nemo")

text = "Oli otya?"
parsed = fastpitch.parse(text)
spec = fastpitch.generate_spectrogram(tokens=parsed)
audio = hifigan.convert_spectrogram_to_audio(spec=spec)
sf.write("output.wav", audio.cpu().numpy()[0], 22050)
```

## Training
- Dataset: Sunbird/salt Luganda studio recordings
- Steps: 5,000 (FastPitch fine-tuning)
- Test Results: 8/8 phrases passed