🎵

Audio Models - Text to Speech (Qwen3)

🛠

On-Device Text-to-Speech Voiceover (Korean & English)

Qwen3-TTS: on-device text-to-speech voiceover for Korean and English. Type text, get a spoken audio file back — with a picked narrator voice preset. Runs locally in the studio.

What it does

Qwen3-TTS turns written text into spoken audio. Give it a sentence, paragraph, or script in Korean or English, choose a narrator voice preset, and it produces a voiceover audio clip. Unlike commercial TTS services (ElevenLabs, Google Cloud TTS, OpenAI TTS), this node runs entirely on-device — the audio never leaves your machine, and you don't need an API key or subscription. Qwen3-TTS is developed by Alibaba's Qwen team as part of their broader Qwen3 speech stack (which also includes Qwen3-ASR for speech recognition). The upstream model supports voice design, voice cloning, and voice presets; the studio exposes the preset flow (pick a named voice) as the first option, keeping the workflow simple.

Problem it solves

  • On-device voiceover – Generate spoken audio without sending text to a third-party API — good for privacy-sensitive scripts, offline workflows, or when you don't want to manage API keys and quotas
  • Korean + English support – Bilingual out of the box, so you can voice both language markets from one node
  • Auto language handling – By default the model detects the language from the text; you can override to force Korean or English if the model gets it wrong on mixed-language input
  • Narrator voice presets – Pick a named voice preset and get a consistent narrator persona across generations
  • Free – No per-minute API cost, no monthly subscription, no rate limits — just local compute
  • Script narration – Turn long-form written content (blog posts, product descriptions, video scripts) into voiceovers
  • Video / animation dubbing – Generate narration tracks for videos or animations without a voice actor
  • Accessibility – Convert written content into audio for users who prefer to listen

Input/Output

  • Input: Text to speak
    • Plain text in Korean or English (or a mix — the model can handle bilingual text)
  • Options:
    • Narrator voice preset (default: default)
      • Select a named voice preset (a specific narrator identity — timbre, pitch, and speaking style)
      • The upstream Qwen3-TTS model also supports voice design and voice cloning; the studio node currently exposes the preset flow
    • Advanced Options:
      • Spoken language hint (Auto / Korean / English, default Auto)
        • Auto — model detects the language from the input text itself (recommended, especially for mixed text)
        • Korean — force Korean pronunciation and prosody
        • English — force English pronunciation and prosody
  • Output: An audio file containing the spoken voiceover of the input text, in the selected voice

Accuracy & Speed

Type
Local (on-device) model — no external API dependency
Provider
Alibaba Qwen — part of the Qwen3 speech stack (siblings: Qwen3-ASR, Qwen3-ForcedAligner)
Languages
Korean and English (UI-exposed); the upstream model may support additional languages
API key required?
No — runs locally in the studio
Internet required?
No — audio generation happens on-device
Voice control
Preset selection in the studio (upstream model supports voice design and voice cloning as well)
Compare to external TTS
No per-minute cost or API quota; audio stays on-device (privacy); trade-off is local compute usage vs. offloading to a cloud API

Technical Details

Model family
Qwen3-TTS (Alibaba Qwen — Qwen3 speech stack)
Task
Text-to-Speech (neural voiceover)
Sibling models in the Qwen3 speech stack
Qwen3-ASR-0.6B-hf / Qwen3-ASR-1.7B-hf (speech recognition), Qwen3-ForcedAligner-0.6B-hf (speech-text alignment)
Full capabilities of the upstream model
Voice design (create a voice from a description), voice cloning (mimic a reference voice sample), voice presets (pick a named voice) — the studio surfaces the preset flow
Deployment style
On-device inference — no external API endpoint required
Languages
Korean, English (UI-exposed) — upstream model may support additional languages, refer to the Qwen3-TTS demo for the full list

Compliance & Provenance

Provider
Alibaba (open weights)
Provider type
Specialized
License
Apache 2.0
EU AI Act risk class
Limited Risk
Art. 50 transparency
Required — outputs are marked. See AI Policy §2.
Region availability
Available globally
Training data summary
Provider model card / training data

For more on how we classify models and mark outputs, see our AI Policy.

Model Source