Qwen3-TTS

Freehand’s Qwen3-TTS 1.7B CustomVoice profile uses Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice through vLLM-Omni v0.18.0.

Choose the model and voice

1. Follow the vLLM-Omni connection guide.
2. Open Text to speech → Settings → Speech. Select the connection and served model, then choose Qwen3-TTS 1.7B CustomVoice as the model profile.
3. Turn on Enable text to speech and choose a preset voice.

The voice picker offers the checkpoint’s nine preset speakers: Vivian, Serena, Uncle Fu, Dylan, Eric, Ryan, Aiden, Ono Anna, and Sohee. Refresh voices reads server metadata without generating audio.

Shape the delivery

Control	Choices or effect
Speech language	Automatic, Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, or Italian
Voice style	Up to 500 characters describing delivery; leave empty for the voice’s usual style
Speaking speed	Requests a speed from the server; does not change playback locally

Describe the delivery

For example: “Speak warmly and clearly, with a relaxed pace.”

Choose Preview to hear the current edits before saving. The preview uses your draft voice, model profile, language, style, speed, and timeout together. It does not save those edits. Choose Save to apply them to the composer and explicit Listen actions.

Language and style are remembered with this model’s voice settings. Switching models keeps speaking speed in place. Generated audio stays in memory until you clear it, replace it, start recording, or quit. You can explicitly save a WAV file; saving does not clear the audio from memory.

Use the matching checkpoint

This profile supports 1.7B CustomVoice. Qwen’s Base voice-cloning, VoiceDesign, and 0.6B variants are outside this contract.

Generate and play speechUse the composer, listen to transcripts, control playback, or save a WAV.