Qwen3-TTS Text to Speech with Preset Voices You Can Direct
Last verified: October 29, 2026
Qwen3-TTS reads the meaning of a script and adjusts tone, speaking rate and emotion to fit it. It also follows plain-language instructions about how the lines should be delivered. With Qwen3-TTS text-to-speech on Vidofy, you paste the script into the Text box and choose a speaker under Voice. When you want a particular read, you describe it in Style: "calm and unhurried" for a tutorial, for example, or "breathless and excited" for a launch teaser.
The speakers are Qwen's preset premium timbres, picked to cover different mixes of gender, age, language and dialect. The model covers ten major languages, and every speaker can speak all of them, even though each has a native one. So each control has one clear job: Voice decides who is speaking, Language tells the model what it is reading, and the script plus the Style note decide how it sounds.
Explore more text to speech
Qwen3-TTS Form at a Glance
What goes into each run and what comes back out.
Input
A text prompt of up to 5,000 characters
Voice
9 options
Language
11 options
Output
1 audio per run
How Do You Get the Most Natural Qwen3-TTS Voice?
Pair Each Speaker with Their Native Language
Set the Language Menu Once You Know the Script's Language
How Do I Turn a Script into Speech with Qwen3-TTS?
Four steps take you from a pasted script to a downloaded voice track.
Write your script in the "Text" box
Type or paste the words you want spoken, up to 5,000 characters. This box is required. Keep directions like "whisper this part" out of it and put them in "Style" instead, because everything in this box is treated as script to be spoken. If you're starting from rough notes, "Enhance Prompt" under the box can suggest a fuller draft.
Choose "Voice" and "Language"
Under "Voice", pick Aiden, Dylan, Eric, Ono Anna, Ryan, Serena, Sohee, Uncle Fu or Vivian; Serena is selected by default. "Language" starts on Auto, which detects the language from your text. It also offers Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian and Spanish. For a particular read, add up to 200 characters in "Style" describing tone, pace or emotion. Style is optional, so you can hear the plain read first and add direction afterwards.
Click "Generate"
Before you click "Generate", check that the speaker, the language and the "Style" note fit the script you pasted, since a mismatch means generating again. A short test passage is a quick way to audition a speaker before you commit the full script.
Download the result
Each run returns one audio result. Listen all the way through, paying attention to names, figures and the closing lines, then download it. If the read isn't what you wanted, change the speaker or the "Style" note and generate again.
Generate Here or Call Qwen's Speech API?
How the form compares with using Qwen speech models through Alibaba Cloud Model Studio.
| Criterion | Qwen3-TTS | Alternatives | Best For |
|---|---|---|---|
| Getting started | You paste a script, choose a speaker and a language from menus and click Generate, with no code to write. | The Qwen-TTS API on Alibaba Cloud Model Studio is called from your own code through the DashScope SDK and requires a Model Studio API key. | The form suits creators who want a finished voice track quickly, while developers building speech into an app or batch pipeline need the API. |
| Directing the delivery | Every run can carry a short Style note on tone, pace or emotion alongside the speaker you picked. | Model Studio offers instruction control on its Qwen3-TTS Instruct-Flash models, while the standard Qwen3-TTS-Flash models do not support it. | You can direct the read on either side, but API users have to choose an instruct-capable model to do it, while form users just fill in Style. |
| Voice roster | The Voice menu offers the Qwen3-TTS preset speakers, each with its own described timbre and native language. | Qwen3-TTS-Flash on Model Studio is documented with 17 expressive voices. | If one of the form's presets fits your project, stay with it; if you need a timbre none of them has, look at the API's roster. |
| Live output | Each run delivers a finished audio result that you play back and download. | Model Studio supports real-time speech synthesis through streaming calls, sending audio back piece by piece as the text arrives. | Voice agents and live apps that have to speak while text is still arriving belong on the streaming API, while voiceovers, narration and social clips fit the form. |
Common Qwen3-TTS Problems and Fixes
What to change when a run doesn't give you the audio you planned.
My script doesn't fit in the Text box.
My Style note doesn't seem to change how the line sounds.
I need two characters talking, but the result has only one voice.
Frequently Asked Questions
Can Qwen3-TTS read messy or complex text?
Yes. The model is designed to handle complex text and is much more robust to noisy input, so you can run a draft copied from a document or web page for a first listen before you spend time cleaning it up. Listen to the whole result and edit only the passages that come out wrong.
Will Enhance Prompt rewrite my script?
Only if you accept one of its suggestions. Enhance Prompt sits under the Text box and suggests fuller versions of a rough idea; your text stays exactly as you wrote it until you pick one. Use it to turn a few bullet points into a draft that reads well aloud, then read the version you chose all the way through, because everything in the Text box gets spoken.
Do I have to fill in the Style field?
No, it's optional. Even without a note, Qwen3-TTS adjusts tone, rhythm and emotion to match what the text says. Start with a plain run, and add Style only when you want a read the words don't suggest by themselves, such as a deadpan delivery of an excited line.
What makes one generation cost more than another?
The credit cost of a run is calculated from what you submit and shows above the Generate button once you're signed in. It changes with the length of your prompt, meaning the text in the Text box, so a short line and a full script are priced differently. Enhance Prompt suggestions don't use credits.
Can I redo one sentence without regenerating the whole script?
Yes. Paste just that sentence, or the short paragraph around it, into the Text box, set Voice, Language and Style exactly as they were for the first run, and generate again. Then put the new clip in place of the old one in your editor. Keeping every setting the same gives the replacement the best chance of blending in.