> ## Documentation Index > Fetch the complete documentation index at: https://docs.goenhance.ai/llms.txt > Use this file to discover all available pages before exploring further. # Qwen3 TTS Voice Clone > Clone a voice from a short recording and have it read any text. The reference and the text don't need to share a language — an English recording can read Chinese, and vice versa. Speaks Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish and Italian. Returns an **mp3** (default) or **WAV** URL (24 kHz, mono). There is no enrollment step and no voice ID to manage: send the reference recording with each request. **For the best clone**, use 5–20 seconds of clear speech from a single speaker, without background music or noise — whatever is in the recording (other voices, music, room echo) tends to carry over. The pace follows the reference, so there is no `speed` parameter. **Pricing** — billed by text length, not by audio duration: | Unit | Tokens | USD | |---|---|---| | every 200 characters (rounded up) | 1 | $0.02 | A 350-character request is billed as 2 units (2 tokens = $0.04). Anything from 1 to 200 characters costs 1 unit. Returns an `img_uuid`; poll GET /api/v1/jobs/detail (or use `custom_callback_url`) to get the generated audio. The result arrives as a `json` array item with `type: "audio"` and the audio URL in `value`. If generation fails, the task ends as `failed` and the tokens are refunded automatically. ## OpenAPI ````yaml api-reference/audio-generations/openapi-qwen3-tts-voice-clone.json POST /api/v1/audio/generations openapi: 3.1.0 info: title: GoEnhance API - Qwen3 TTS Voice Clone description: Qwen3 TTS Voice Clone version: 1.0.0 servers: - url: https://api.goenhance.ai security: [] paths: /api/v1/audio/generations: post: tags: - AudioGenerations summary: Qwen3 TTS Voice Clone description: >- Clone a voice from a short recording and have it read any text. The reference and the text don't need to share a language — an English recording can read Chinese, and vice versa. Speaks Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish and Italian. Returns an **mp3** (default) or **WAV** URL (24 kHz, mono). There is no enrollment step and no voice ID to manage: send the reference recording with each request. **For the best clone**, use 5–20 seconds of clear speech from a single speaker, without background music or noise — whatever is in the recording (other voices, music, room echo) tends to carry over. The pace follows the reference, so there is no `speed` parameter. **Pricing** — billed by text length, not by audio duration: | Unit | Tokens | USD | |---|---|---| | every 200 characters (rounded up) | 1 | $0.02 | A 350-character request is billed as 2 units (2 tokens = $0.04). Anything from 1 to 200 characters costs 1 unit. Returns an `img_uuid`; poll GET /api/v1/jobs/detail (or use `custom_callback_url`) to get the generated audio. The result arrives as a `json` array item with `type: "audio"` and the audio URL in `value`. If generation fails, the task ends as `failed` and the tokens are refunded automatically. parameters: - name: Authorization in: header description: '' required: false example: '{{Authorization}}' schema: type: string requestBody: content: application/json: schema: type: object properties: model: type: string enum: - qwen3-tts-voice-clone description: Model name. Must be `qwen3-tts-voice-clone`. text: type: string minLength: 1 maxLength: 5000 description: >- Text to synthesize. Required. Max 5000 characters. Billing is based on this length. Long texts are read in sentence-sized parts and joined into one file, so a single request can cover a whole article. reference_audio_url: type: string format: uri description: >- Required. Public HTTPS URL of the voice to clone. Max 50MB. Common audio formats (mp3, wav, m4a, aac, ogg, flac) and video files (the audio track is used) are accepted. Needs at least 2 seconds of speech. Unreachable or oversized links are rejected before any tokens are charged; if the file turns out not to be audio or contains no speech, the task fails and is refunded. example: https://your-cdn.com/voice-sample.mp3 reference_text: type: string maxLength: 1000 description: >- Optional. The exact words spoken in the reference recording. When given, it must match the recording word for word, and the whole recording is used (up to 60 seconds). When omitted, the first 20 seconds or less (cut at a natural pause) are used and transcribed automatically. language: type: string enum: - auto - chinese - english - japanese - korean - german - french - russian - portuguese - spanish - italian - zh - en - ja - ko - de - fr - ru - pt - es - it default: auto description: >- Language of `text`. Optional — defaults to `auto`. Case-insensitive; the two-letter codes are accepted too. Supported: Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish and Italian. Setting it explicitly when you know the language gives the most stable result. output_format: type: string enum: - mp3 - wav default: mp3 description: >- Audio format. Optional — `mp3` (default, 128 kbps) or `wav` (16-bit). Both are 24 kHz mono. custom_callback_url: type: string format: uri description: >- Optional. A publicly accessible HTTPS URL. When the task status changes (processing / success / failed), GoEnhance sends a POST request to this URL. The request body is identical to the response of GET /api/v1/jobs/detail. If your server does not respond with HTTP 200, the notification is retried up to 3 times, with a 3-second timeout per attempt. example: https://your-server.com/goenhance/callback required: - model - text - reference_audio_url example: model: qwen3-tts-voice-clone text: This voice was cloned from a short recording. reference_audio_url: https://your-cdn.com/voice-sample.mp3 language: english responses: '200': description: '' content: application/json: schema: type: object properties: code: type: integer msg: type: string data: type: object properties: img_uuid: type: string cost: type: number description: >- Tokens deducted for this request. This is the amount actually charged, so it already reflects any discount active on your account and can be lower than the listed price. Tokens are deducted when the task is accepted, and refunded automatically if the generation ends in failure. example: 1 required: - img_uuid - cost required: - code - msg - data examples: '1': summary: Success value: code: 0 msg: Success data: img_uuid: c12b656c-747a-44fd-9c80-add79b0c52d5 cost: 1 '2': summary: Insufficient tokens value: code: 100 msg: tokens is not enough quota: summary: API key quota exceeded value: code: 101 msg: 'API key quota exceeded: 4990 of 5000 tokens used (monthly)' headers: {} '401': description: '' content: application/json: schema: type: object properties: code: type: integer msg: type: string headers: {} deprecated: false ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.