Qwen 3 TTS Text To Speech 1 7b
Convert text into speech using a selected voice or a supplied voice profile, with language, style, and advanced voice controls.
Pricing
108 credits per 1,000-character billing unit. Successful requests with fewer than 1,000 characters are charged for one unit. The final charge is based on the number of billing units used.
Charges apply only to successful outputs; server errors and queue waiting time are not charged. This covers synthesis with preset voices or an existing speaker embedding; creating a cloned voice configuration is outside this tool's scope.
Parameters
Optional parameters use the platform defaults when omitted.
| Parameter | Type | Required | Description | Default |
|---|---|---|---|---|
request_body | object | Yes | No parameter description. | — |
request_body.text | string | Yes | The text to be converted to speech. | — |
request_body.top_k | integer | unknown | No | Top-k sampling parameter. | 50 |
request_body.top_p | number | unknown | No | Top-p sampling parameter. | 1 |
request_body.voice | string | unknown | No | The voice to be used for speech synthesis, will be ignored if a speaker embedding is provided. Check out the **[documentation](https://github.com/QwenLM/Qwen3-TTS/tree/main?tab=readme-ov-file#custom-voice-generate)** for each voice's details and which language they primarily support. | — |
request_body.prompt | string | unknown | No | Optional prompt to guide the style of the generated speech. This prompt will be ignored if a speaker embedding is provided. | — |
request_body.language | enum (Auto | English | Chinese | Spanish | French | German | Italian | Japanese | Korean | Portuguese | Russian) | No | The language of the voice. | Auto |
request_body.temperature | number | unknown | No | Sampling temperature; higher => more random. | 0.9 |
request_body.max_new_tokens | integer | unknown | No | Maximum number of new codec tokens to generate. | 200 |
request_body.reference_text | string | unknown | No | Optional reference text that was used when creating the speaker embedding. Providing this can improve synthesis quality when using a cloned voice. | — |
request_body.subtalker_top_k | integer | unknown | No | Top-k for sub-talker sampling. | 50 |
request_body.subtalker_top_p | number | unknown | No | Top-p for sub-talker sampling. | 1 |
request_body.repetition_penalty | number | unknown | No | Penalty to reduce repeated tokens/codes. | 1.05 |
request_body.subtalker_dosample | boolean | unknown | No | Sampling switch for the sub-talker. | true |
request_body.subtalker_temperature | number | unknown | No | Temperature for sub-talker sampling. | 0.9 |
request_body.speaker_voice_embedding_file_url | string | unknown | No | URL to a speaker embedding file in safetensors format, from `fal-ai/qwen-3-tts/clone-voice` endpoint. If provided, the TTS model will use the cloned voice for synthesis instead of the predefined voices. | — |
Invocation example
Copy the request and explore its response example.
/agent-api/v1/tool/executeX-Api-KeySubmit first, then query with the returned task_id until the task completes.
curl --request POST \
--url 'https://agentearth.ai/agent-api/v1/tool/execute?tool_name=ae_qwen_3_tts_text_to_speech_1_7b' \
--header "X-Api-Key: $AGENTEARTH_API_KEY" \
--header 'Content-Type: application/json' \
--data-raw '{
"params": {
"request_body": {
"language": "Chinese",
"max_new_tokens": 512,
"prompt": "憨厚可爱的卡通胖猫拟人声线:圆滚滚、慢吞吞,说话有点含糊,鼻音偏重,憨态可掬,像一头老实又懒洋洋的大胖熊,尾音往下掉。",
"text": "嗯。这事儿,不归我管。……行了,放那儿吧。我没说不管。",
"voice": "Eric"
}
}
}'{
"description": "task accepted; poll the result via query_tool with this task_id",
"query_tool": "ae_get_result",
"status": "created",
"task_id": "aac3dad2-2412-4ec6-84b2-fc753f7cb502"
}This example shows part of the response. Copied content is valid JSON.
Guides using this tool
See how this tool fits into real workflows





