gafiatulin / vibevoice-tts-0.5b-coreml

VibeVoice Realtime 0.5B (Qwen2.5-0.5B) — CoreML streaming TTS, 25 built-in voices across 10 languages.

Usage

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

import VibeVoiceCoreML

let tts = try await RealtimeTTS()
for try await frame in tts.speak("Hello world", config: TTSConfig(voice: "Emma")) {
    // 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

• base_lm_stateful.mlmodelc
• tts_lm_stateful.mlmodelc
• diffusion_head_b2.mlmodelc
• vae_decoder_streaming.mlmodelc
• eos_classifier.mlmodelc
• acoustic_connector.mlmodelc

Data

• embed_tokens.bin
• tts_input_types.bin

Extras

• tokenizer.json
• tokenizer_config.json
• voices/de-Spk0_man.vvvoice
• voices/de-Spk1_woman.vvvoice
• voices/en-Carter_man.vvvoice
• voices/en-Davis_man.vvvoice
• voices/en-Emma_woman.vvvoice
• voices/en-Frank_man.vvvoice
• voices/en-Grace_woman.vvvoice
• voices/en-Mike_man.vvvoice
• voices/fr-Spk0_man.vvvoice
• voices/fr-Spk1_woman.vvvoice
• voices/in-Samuel_man.vvvoice
• voices/it-Spk0_woman.vvvoice
• voices/it-Spk1_man.vvvoice
• voices/jp-Spk0_man.vvvoice
• voices/jp-Spk1_woman.vvvoice
• voices/kr-Spk0_woman.vvvoice
• voices/kr-Spk1_man.vvvoice
• voices/nl-Spk0_man.vvvoice
• voices/nl-Spk1_woman.vvvoice
• voices/pl-Spk0_man.vvvoice
• voices/pl-Spk1_woman.vvvoice
• voices/pt-Spk0_woman.vvvoice
• voices/pt-Spk1_man.vvvoice
• voices/sp-Spk0_woman.vvvoice
• voices/sp-Spk1_man.vvvoice

License

MIT (same as upstream VibeVoice models from Microsoft)