| { | |
| "model_name": "TruthShield VoiceGen", | |
| "model_type": "vits", | |
| "multilingual": true, | |
| "supported_languages": [ | |
| "bhojpuri", | |
| "bengali", | |
| "english", | |
| "gujarati", | |
| "hindi", | |
| "chhattisgarhi", | |
| "kannada", | |
| "magahi", | |
| "maithili", | |
| "marathi", | |
| "telugu" | |
| ], | |
| "speaker_embedding_model": "ECAPA-TDNN", | |
| "vocoder": "HiFiGAN", | |
| "sampling_rate": 22050, | |
| "description": "Configuration reference for the TruthShield VoiceGen multilingual, multi-speaker TTS system. Fine-tuned weights are stored externally for evaluation as required by the Voice Tech For All Challenge.", | |
| "architecture": { | |
| "text_encoder": "VITS Text Encoder", | |
| "duration_predictor": "VITS Duration Module", | |
| "decoder": "Flow-based VITS Decoder", | |
| "prosody_features": ["pitch", "energy"], | |
| "speaker_embedding_dim": 192 | |
| }, | |
| "safety_layer": { | |
| "enabled": true, | |
| "similarity_model": "ECAPA-TDNN", | |
| "outputs": ["similarity_score", "safety_verified"] | |
| }, | |
| "api_spec_compliance": { | |
| "endpoint": "GET /Get_Inference", | |
| "mandatory_params": ["text", "lang", "speaker_wav"] | |
| } | |
| } | |