Modelpublic

vaghawan/xtts-v2-stage-c-balanced-dual-voice-3epoch-1h-bible_tts_speaker-hausa-speaker-female-waxalnlp-3

sourceHugging Faceotherupdated 26d agoView on Hugging Face

0likes16downloads

Model Card

XTTS-v2 Hausa — Stage C — balanced BibleTTS + waxalnlp dual-voice (3 epochs, from Stage A)

Fine-tuned Coqui XTTS-v2 for Hausa (ha), trained on Balanced OpenSLR BibleTTS + vaghawan/hausa-tts-24khz-waxalnlp-3-clean (~441 clips each), init from Stage A checkpoint.

Files

File	Purpose
`best_model.pth`	Fine-tuned GPT checkpoint (epoch 3)
`config.json`	Model config (includes Hausa language)
`vocab.json`	Extended Hausa BPE vocabulary
`references/bible_tts_speaker.wav`	Speaker reference for `bible_tts_speaker`
`references/hausa-speaker-female-waxalnlp-3.wav`	Speaker reference for `hausa-speaker-female-waxalnlp-3`
`infer.py`	Inference script
`xtts_hausa_patch.py`	Required Hausa runtime patches
`env_config.py`	Config loader
`config.env.example`	Example settings (copy to `config.env`)
`samples.txt`	Default eval sentences
`requirements.txt`	Python dependencies

Quick start

bash

pip install -r requirements.txt
cp config.env.example config.env

python infer.py --model-dir . --speaker-wav references/bible_tts_speaker.wav

# Single line:
python infer.py --model-dir . --text "Ina kwana." --speaker-wav references/bible_tts_speaker.wav --out outputs/out.wav

All speakers (one WAV per reference):

python infer.py --model-dir . --all-speakers --out-dir outputs/samples

Training details

—Base model: Coqui XTTS-v2
—Epochs: 3
—Language: Hausa (ha)
—Speakers: bible_tts_speaker, hausa-speaker-female-waxalnlp-3

License

XTTS-v2 uses the Coqui Public Model License. Check each dataset license before redistribution.