gafiatulin / vibevoice-tts-7b-coreml

VibeVoice 7B (Qwen2.5-7B) — CoreML INT8, fused LM+head, fused diffusion loop, DPM-Solver++ 10-step. Multi-speaker TTS with voice cloning.

Usage

Add vibevoice-coreml to your Swift package. Models auto-download from this repo on first use.

import VibeVoiceCoreML

let tts = try await MultispeakerTTS(architecture: .model7B)
let voices = try await tts.encodeVoices(from: [referenceAudioURL])
for try await frame in tts.speak("Hello world", config: MultispeakerConfig(), voices: voices) {
    // frame.samples: [Float] at 24kHz
}

See the GitHub repo for CLI usage, Python pipelines, and conversion scripts.

Requirements

• macOS 15+ (requires ct.StateType for stateful models)
• Pre-compiled .mlmodelc — no on-device compilation needed

Files

Models

• lm_decoder_fused_int8.mlmodelc
• diffusion_loop.mlmodelc
• vae_decoder_streaming.mlmodelc
• semantic_encoder_streaming.mlmodelc
• acoustic_connector.mlmodelc
• semantic_connector.mlmodelc
• vae_encoder.mlmodelc

Data

• embed_tokens.bin

Extras

• tokenizer.json
• tokenizer_config.json

License

MIT (same as upstream VibeVoice models from Microsoft)