Index-Echo-S2ST-9B-MLX
Native Apple Silicon (MLX) optimized weights for Index-Echo-S2ST-9B, supporting speech-to-speech translation and video dubbing with original speaker voice preservation.
GitHub Repository: vanch007/video-dub-mlx
Pure MLX Model Architecture
- stlm_llm/: 9B Speech Translation Language Model in pure MLX native tensor format (language_model.model.layers.*).
- bridge/: Hidden2CV Cross-Modal Bridge Mapper in native MLX format (mapper.safetensors).
- stlm_ckpt/: Audio Tower Conformer weights (model.safetensors).
- cosyvoice3_mlx/: Pure MLX CosyVoice3 acoustic model (zero ONNX).
- s3tokenizer_mlx/: Pure MLX S3Tokenizer V3 discrete speech codec.