Use Qwen Audio 3.1 TTS Flash through EvoLink’s unified API: 68 voices, voice cloning and design. Check model IDs, token pricing and async integration.
Turn text into speech with a system voice, or switch the model to Voice Enrollment to clone or design your own voice and use it right away.
Create speech task
Fixed to qwen-audio-3.1-tts-flashUp to 5,000 characters. Add tags such as [excited] or [laughing] in the text to shape the delivery.
Pick a system voice from the list, or type or paste a voice name here, including one you created. Voice names are case-sensitive. Pick a voice that supports the language of your text.
Optional. Controls emotion, pace or dialect. Up to 100 billing characters (Han characters, including Japanese kanji, count as 2; kana, Hangul and all other characters count as 1).
Optional hint for the language of the text.
Pitch also changes the duration, and has no effect when speed is not 1.0.
Your audio will appear here.
Speech synthesis is billed by the tokens actually used, with separate input and output rates. Creating a custom voice is a small flat fee per voice. Failed tasks are refunded in full.
First Test Cost
One short sentence · 14 input + 33 output tokensOne short sentence
About 158,139 sentences like this with $10 in credits.
In Playground, keep the default voice and enter one sentence. Credits are reserved from the text length when the task is created, then settled to the tokens actually used: any excess is returned, and a shortfall is charged.
Qwen Audio 3.1 TTS Flash testing budget guide
Pick an amount based on how many tests you expect.Good for a first validation.
Good for narrating a batch of articles or lessons.
Good before production integration.
Token counts are measured samples. Actual usage depends on the language, the voice and how the text is read: digits, upper-case letters and symbols are read out one by one and use more output tokens.
qwen-audio-3.1-tts-flash| Billing item | What it covers | Rate | Billing |
|---|---|---|---|
| Output tokens | The generated speech. Longer audio uses more output tokens. | $1.765/1M tokens120 Credits | By actual usage |
| Input tokens | The text you send. The style instruction is not counted. | $0.221/1M tokens15 Credits | By actual usage |
The generated speech. Longer audio uses more output tokens.
The text you send. The style instruction is not counted.
voice-enrollment| Billing item | What it covers | Rate | Billing |
|---|---|---|---|
| Create a voice | Voice cloning (audio_url) or voice design (voice_prompt), same price. The voice is bound to qwen-audio-3.1-tts-flash. | $0.0001/voice0.001 Credits | Per voice created |
Voice cloning (audio_url) or voice design (voice_prompt), same price. The voice is bound to qwen-audio-3.1-tts-flash.
Billing notes
Qwen Audio 3.1 TTS Flash turns text into natural speech with 68 system voices and instruction-based control over emotion, pace and dialect. On the same page, Voice Enrollment creates your own voice for it: clone a consenting speaker from a short recording, or design a new voice from a text description. EvoLink exposes both as asynchronous tasks on one audio endpoint with one API key.
qwen-audio-3.1-tts-flashText-to-speech. Send prompt and an optional voice; get an audio file back. Billed by input and output tokens.
voice-enrollmentCreate a custom voice for Qwen Audio 3.1 TTS Flash. Send audio_url to clone a voice from a recording, or voice_prompt and preview_text to design one from a description. Flat fee per voice.
Speech synthesis parameters first, then the fields used by voice-enrollment. Open the API tab for the full request and response schema.
promptstringDefault: required
Text to synthesize, up to 5,000 characters. Emotion tags such as [excited] can be written inline.
voicestringDefault: longanhuan_v3.1
A system voice (case-sensitive), or a voice your account created with voice-enrollment.
instructionstringDefault: none
Natural-language control of emotion, pace or dialect. Up to 100 billing characters.
response_formatenumDefault: mp3
mp3, wav or opus. Opus supports 8, 12, 16, 24 and 48 kHz only.
speech_rate / pitchnumberDefault: 1.0
Both range from 0.5 to 2.0. Pitch also changes the duration and is ignored when speech_rate is not 1.0.
audio_urlstring · voice-enrollmentDefault: required for cloning
Public HTTP(S) link to a WAV, MP3 or M4A sample with clear speech, up to 60 s and 10 MB. Selects voice cloning.
voice_prompt + preview_textstring · voice-enrollmentDefault: required for design
Voice description up to 500 characters, and a 15–200 character sentence for the preview clip. Selects voice design.
preferred_namestring · voice-enrollmentDefault: required
1–10 English letters or digits. It becomes part of the returned voice name.
Use the EvoLink API for product integration and batch jobs, or call it from Codex, Claude, or Gemini for fast creative and development workflows. Both paths share the same EvoLink API key, balance, model routes, and task history.
Best for: product backends, batch jobs, automated workflows
Call EvoLink’s unified voice API from your server and control the model ID, parameters, task queue, callbacks, and result storage.
Best for: creative and development tasks in Codex, Claude, and Gemini
Give the Agent your output goal, assets, and acceptance criteria. It can choose the route, assemble the request, track the task, and return the result without requiring you to hand-code every step.
This example shows the shortest runnable flow: create a task, save the returned task ID, then poll or wait for an HTTPS callback. Open the API tab for the complete parameter and response reference.
curl -X POST https://api.evolink.ai/v1/audios/generations \
-H "Authorization: Bearer $EVOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen-audio-3.1-tts-flash",
"prompt": "Hello, and welcome to EvoLink. This is a quick demo of Qwen Audio 3.1 TTS Flash.",
"voice": "Emily_v3.1",
"instruction": "Speak warmly, at a relaxed pace",
"response_format": "mp3"
}'
# Save the returned task ID, then query; the audio link is in results[0]:
curl https://api.evolink.ai/v1/tasks/{task_id} \
-H "Authorization: Bearer $EVOLINK_API_KEY"Check every value against the parameter cards above: ranges, allowed options, asset counts, and file types must match the selected route.
Check the Authorization bearer token and confirm the available balance and task billing in the console.
Review likeness rights, brand assets, readable text, sensitive content, and input-asset compliance.
A Qwen Audio 3.1 TTS Flash task that reaches a final failed state is not charged. Keep the task ID, inspect the final state, and identify whether the cause is validation, review, or execution before retrying.
callback_url must be a public HTTPS URL, not localhost or a private network address. Always keep the task ID as a polling fallback.
| Feature | System voice | Voice cloning | Voice design |
|---|---|---|---|
| Model ID | qwen-audio-3.1-tts-flash | voice-enrollment | voice-enrollment |
| What you provide | A voice name from the list | A speech sample (audio_url) | A voice description + preview text |
| What you get | Audio | A voice name | A voice name + preview clip |
| Setup cost | None | ~$0.0001 per voice | ~$0.0001 per voice |
| Best for | Getting started quickly | Reproducing a specific, consenting speaker | New voices, characters, brand tone exploration |
Send qwen-audio-3.1-tts-flash to synthesize speech and voice-enrollment to create a voice. Both go to POST /v1/audios/generations and return a task ID.
Poll /v1/tasks/{task_id} or use callback_url for the result. Completion time varies with the text and voice operation; no fixed latency is guaranteed. Audio links expire after 24 hours.
A custom voice only works with qwen-audio-3.1-tts-flash and only for the account that created it. Using someone else's voice name returns 404. A voice can be used for 6 hours after it is created.
Unknown parameters, an invalid voice name or over-length text fail fast with 400 instead of reserving credits and failing later.
One API key works across Qwen, Seed Audio, video, image, and LLM models.
Track credits, spending, and model usage in one console.
See whether a task is submitted, processing, completed, or failed, with the real failure reason.
Speech is settled on actual token usage, and credits reserved for a failed task are returned in full.
Doubao Seed Audio 1.0
BytePlus prompt-based audio for voice, dialogue, sound effects, music, and ambience.
View API
This EvoLink route uses asynchronous HTTP tasks, without SSE or WebSocket streaming. Submit to POST /v1/audios/generations, then poll GET /v1/tasks/{task_id} for the audio URL. The provider’s streaming APIs use a different protocol.
No. This page covers the hosted qwen-audio-3.1-tts-flash model. Qwen3-TTS, Qwen Audio TTS-Next and Qwen Audio Realtime are separate models with different IDs, parameters, weights and streaming capabilities.
Use qwen-audio-3.1-tts-flash for speech synthesis and voice-enrollment to create a custom voice. Both are sent to POST /v1/audios/generations.
Speech is billed by the tokens actually used, with separate input and output rates shown live in the Pricing section. Creating a custom voice is a small flat fee per voice. Failed tasks are refunded in full.
The token count is only known after synthesis, so credits are reserved from the text length when the task is created, usually more than the final charge. When it completes, you are billed for the tokens actually used and the excess is returned. Text that is read out character by character, such as strings of digits, letters or symbols, can use more than was reserved; the shortfall is then charged.
Call voice-enrollment with audio_url to clone a voice or with voice_prompt and preview_text to design one, read result_data.voice from the finished task, and pass it as the voice parameter to qwen-audio-3.1-tts-flash. A custom voice can be used for 6 hours after it is created; after that, create a new one.
Voice cloning reproduces a real speaker from a short recording and returns the voice name only. Voice design creates a new voice from a text description and also returns a preview clip. Both cost the same and both work only with qwen-audio-3.1-tts-flash.
Required: WAV, MP3 or M4A, up to 60 s and 10 MB, 16 kHz or higher, with clear speech. For best results, use 10–20 seconds from one speaker, mono audio, pauses no longer than 2 seconds, and no background music or noise.
No. A custom voice can only be used by the account that created it; other accounts get a 404 for the same voice name.
Only your own voice or a voice you have explicit permission to use. Cloning someone's voice without consent is not allowed.
Pass a short instruction such as "speak slowly and gently", or write tags like [excited] directly in the text. The four multilingual voices also speak several Chinese dialects when the instruction asks for one.
Failed tasks are not charged—the reserved credits are returned in full. The task result shows the failure reason.