Hindi VITS TTS (V1 Baseline)
This is an end-to-end Text-to-Speech model for Hindi based on the VITS architecture.
Model Details
- Architecture: VITS (Variational Inference with adversarial learning)
- Language: Hindi (Devanagari)
- Status: Research Baseline (V1)
Training Data
Trained on the IIT Madras IndicTTS Database, specifically the Hindi monolingual subset:
- Speakers: 2 (1 Male, 1 Female)
- Sampling Rate: 22,050 Hz
- Content: Standard Hindi utterances with high-quality studio recordings.
Features
- Handles Devanagari Unicode normalization.
- Number-to-word conversion for Hindi digits.
- Trained on Kaggle GPU environments.
