FluffyBunnies/vibevoice-onnx-v2

VibeVoice TTS - ONNX for Browser

Microsoft's VibeVoice-Realtime-0.5B converted to ONNX for browser inference.

Architecture

VibeVoice uses a two-LLM architecture with diffusion-based vocoding:

1. Text LM (187 MB INT8) - Processes input text to hidden states
2. TTS LM (417 MB INT8) - Generates speech latents autoregressively
3. Diffusion Head (25 MB INT4) - Denoises latents
4. Vocoder (339 MB INT4) - Converts latents to 24kHz audio
5. Acoustic Connector (0.5 MB INT4) - Feeds speech back to TTS LM

Total: ~969 MB quantized (down from 3.8 GB)

Files

• text_lm_int8.onnx - Text encoder
• tts_lm_int8.onnx - Speech token generator
• diffusion_head_int4.onnx - Latent denoiser
• vocoder_int4.onnx - Audio synthesizer
• acoustic_connector_int4.onnx - Speech feedback
• tokenizer.json - Qwen2 tokenizer
• config.json - Model configuration

Usage

See demo.html for browser usage example.

License

Apache 2.0 (same as original VibeVoice model)