aitytech/Qwen3-TTS-12Hz-0.6B-CustomVoice-MLX-4bit

Qwen3 TTS 12Hz 0.6B CustomVoice — MLX 4-bit

MLX 4-bit quantized conversion of Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice for Apple Silicon inference.

Usage

Used by qwen3-asr-swift Qwen3TTS module:

let model = try await Qwen3TTSModel.fromPretrained(
    modelId: TTSModelVariant.customVoice.rawValue
)
let audio = try model.synthesize("Hello!", speaker: "Chelsie")

audio speak "Hello!" --model custom-voice --speaker Chelsie -o output.wav

Model Details

• Architecture: Qwen3-TTS (Talker transformer + Code Predictor + Mimi speech tokenizer decoder)
• Parameters: 0.6B
• Quantization: 4-bit (MLX, talker only)
• Size: ~964 MB
• Sample rate: 24 kHz
• Codec rate: 12.5 Hz
• Voices: 9 preset voices + instruction-based style control