Skip to main content

Qwen3 TTS API

Uncensored — This model declares no content filter. SpicyAPI forwards the request without adding a platform content scan; legal and consent duties remain with the customer.Browse every model at this filtering level
alibaba/qwen-3-tts/text-to-speech

Qwen3 TTS is Alibaba's open-source text-to-speech model: send a script, pick one of nine preset voices and get a 24 kHz mono MP3. Plain-language instructions set tone, emotion or pace, and a language can be specified.

From $0.068/1,000 characters
Run model

Workbench

Input

Text to speech
119/2000
The words to be spoken, exactly as written. Billed per 1,000 characters, minimum 100 per request. Up to 2,000 characters. at least 1 characters · at most 2000 characters

119 characters

One of nine preset voices. Defaults to Serena.
48/1000
Plain-language direction on tone, emotion or pace. Shapes the delivery and is not spoken. Up to 1,000 characters; not billed. at most 1000 characters

Never read aloud. It only shapes how the script is performed.

Language of the reading. auto detects it from the script.
Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

Checking your account before enabling submission and pricing.

Output

Ready for a new run
Example result
0:00 / –:––
Avatar
Speech to text
Lip sync

Related models and next tasks

Missing the model you need?

Tell us what to add and where you would use it. Requests go to the people who decide what gets onboarded next.

Nine voices. Your direction.

Qwen3 TTS is Alibaba's open-source text-to-speech model: send up to 2,000 characters of text, pick one of nine preset voices, and add a sentence of plain-language direction on tone, emotion or pace if you want it. The model detects the language automatically or you can set one of ten yourself, including two Chinese dialect voices. Billing is per 1,000 characters of the text you send, with a 100-character minimum, so the quote shown before the run is the exact charge. SpicyAPI applies no content filter, and in our tests our strongest test scripts were read in full.

Open the playground
Nine voices. Your direction.
Voices
9 presets, including two Chinese dialect speakers
Languages
10, auto-detected or set with language
Characters per request
Up to 2,000; billed per 1,000, minimum 100
Output
24 kHz mono MP3
Pricing
$0.068 per 1,000 characters, quote is final
Content filter
None — our strongest test scripts read in full
Voices
Nine fixed characters, two of them dialect speakersChoose a preset for each character and reuse its voice_id. Listen across clips before publishing: a preset does not guarantee identical delivery on every request.
Direction
Delivery notes, not slidersUse instructions for a short note on tone, emotion or pace. These notes are not spoken or counted as billable text. Each successful generation is billed separately for its script, even when you reuse the same words.
Billing
The quote before the run is the final priceThe character count is known the moment you send the request, so what the quote shows is what gets charged — no metering after the fact, no rounding once the audio comes back. A request under the 100-character minimum is still billed as 100.

What the Qwen3 TTS API does well

  • Nine preset voices for your cast

    Reuse the same voice_id and include the delivery instructions in every request. Review the clips together for changes in tone, pacing or pronunciation.

  • Plain-language direction instead of parameter tuning

    instructions reads like a note to a voice actor: 'breathless and teasing, close to the microphone' or 'flat, tired, reading from a script.' It shapes tone, emotion and pace without exposing separate pitch or speed sliders, and it costs nothing extra to include.

  • Ten languages, one field

    language defaults to auto, which detects the language from the script itself, so a mixed catalog of English, Spanish and Japanese lines needs no per-language routing. Setting the field by hand is there for the case where a short or mixed script gets misread.

  • Native dialect and language voices

    Dylan and Eric read in Beijing and Sichuan dialect rather than a standard-accent approximation of them, and Ono_Anna and Sohee are native Japanese and Korean speakers. For a line that has to sound local rather than merely translated, the voice choice does part of the work.

  • A price you can compute before you send the request

    Billing is linear per 1,000 characters of text with a 100-character minimum, and instructions is not counted at all. Because the character count is known upfront, the estimate shown in the console is the amount charged — there is nothing to reconcile after the audio comes back.

  • Unfiltered scripts read as written

    SpicyAPI adds no content filter on top of the model. In our tests, Qwen3 TTS read our strongest first-person and narrated test scripts word for word rather than skipping or softening them, which matters for uncensored audio fiction and companion-voice products where a filtered line breaks the scene.

Who Qwen3 TTS is for

  • A character voice to reuse across lines

    Character and companion apps

    A character voice to reuse across lines

    Reuse the same voice_id and include the delivery instructions in every request. Review the clips together for changes in tone, pacing or pronunciation.

  • One endpoint for a catalog that mixes languages

    Multilingual content teams

    One endpoint for a catalog that mixes languages

    auto language detection means a script in Japanese, then one in Spanish, then a mixed sentence, all go through the same call without a routing table in front of it. Ono_Anna and Sohee give Japanese and Korean lines a native-sounding read rather than an accented one.

  • Narration that does not stop halfway through a scene

    Audiobook and fiction producers

    Narration that does not stop halfway through a scene

    In our tests, our strongest monologue and narration scripts were read in full, word for word. For fiction publishers whose catalog includes that material, a narration step that keeps going through the scene is what makes the workflow usable at all.

  • A voice track before the visuals

    Video and short-form creators

    A voice track before the visuals

    Generate the line here as an MP3, then hand it to a talking-avatar or lip-sync model so a face performs it, or drop it straight onto a video timeline. Per-1,000-character pricing keeps a short caption line and a long narration comparable in cost per word.

How to call the Qwen3 TTS API

  1. Write the script in text

    Up to 2,000 characters, punctuated the way you want it paced. Anything under 100 characters is billed as 100, so a short line does not need padding for cost reasons, but it will not be cheaper than a 100-character one.

  2. Pick a voice_id

    Nine presets are available, defaulting to Serena, a warm young female voice. Ryan and Aiden are English male voices, Dylan and Eric read Beijing and Sichuan dialect, and Ono_Anna and Sohee are native Japanese and Korean voices.

  3. Add instructions if the delivery needs shaping

    A sentence or two of plain language — tone, emotion, pace — steers the reading. It is optional, not spoken aloud, and does not add to the bill.

  4. Set language only if auto gets it wrong

    Leave language on auto for most scripts; the model detects it from the text. Set it explicitly for a short or mixed-language script where automatic detection has less to go on.

  5. Check the quote, then read the MP3 off the result

    The quote shown before the run is the final price, since the character count is already known. The task returns one 24 kHz mono MP3, kept in SpicyAPI storage for 14 days.

Qwen3 TTS worked examples

  • A companion character with a fixed voice

    Reuse voice_id and instructions for each line, then listen to the clips together before publishing.

    {"text": "You came back. I wasn't sure you would, not after last time.", "voice_id": "Vivian", "instructions": "breathless and a little teasing, close to the microphone"}

  • A Korean line read by a native voice

    Setting voice_id to Sohee gives a Korean script a native-sounding read instead of an accented one.

    {"text": "오랜만이에요. 오늘 밤은 그냥 당신 목소리가 듣고 싶었어요.", "voice_id": "Sohee", "language": "ko"}

  • A late-night audio note

    A short personal-sounding line with a delivery note, defaulting to Serena's voice.

    {"text": "It's almost midnight and I still can't stop thinking about you.", "instructions": "soft, low volume, like she doesn't want anyone else in the room to hear"}

Qwen3 TTS API questions

Choose a voice_id for each character and reuse it across the series. Keep instructions short and specific about mood and pace; the direction itself is not billed, but each successful generation is. Leave language on auto unless a short or ambiguous line is detected incorrectly. Listen to each result before publishing.