# Qwen3 TTS - Core.Today AI API

> Alibaba's unified TTS with three modes: preset speakers (custom_voice), instant voice cloning from reference audio (voice_clone), and creating a brand-new voice from a text description (voice_design).

- **Provider**: Qwen
- **Model ID**: qwen/qwen3-tts
- **Category**: Audio & TTS
- **Credits**: 47 per 1000 characters (0.045 credits/char, billed per character)
- **Speed**: Fast
- **Quality**: High

## Features
- Three modes: preset voice, voice cloning, voice design
- Voice design from natural-language description
- 9 preset speakers including Korean (Sohee)
- 10 language support with auto detection
- Style/emotion instruction support

## Use Cases
- Custom voice creation without recordings
- Voice cloning for personalization
- Multilingual assistant voices
- Character voice design for games
- Style-directed narration

## API Endpoint

Base URL: https://api.core.today/v1
Create Prediction: POST /predictions
Get Status: GET /predictions/{job_id}

## Authentication
Header: X-API-Key: YOUR_API_KEY

## Input Parameters

### Required
- **text**: string - Text to synthesize into speech

### Optional
- **mode**: string (default: custom_voice) - 'custom_voice' uses preset speakers, 'voice_clone' clones from reference audio, 'voice_design' creates a voice from a description
  Options: custom_voice, voice_clone, voice_design
- **language**: string (default: auto) - Language of the text ('auto' for automatic detection)
  Options: auto, Chinese, English, Japanese, Korean, French, German, Italian, Spanish, Portuguese, Russian
- **speaker**: string (default: Serena) - Preset speaker voice (custom_voice mode only)
  Options: Aiden, Dylan, Eric, Ono_anna, Ryan, Serena, Sohee, Uncle_fu, Vivian
- **voice_description**: string - Natural-language description of the desired voice (voice_design mode only), e.g. 'A warm, friendly female voice with a slight British accent'
- **reference_audio**: string - Reference audio for voice cloning (voice_clone mode only)
- **reference_text**: string - Transcript of the reference audio (recommended for voice_clone mode)
- **style_instruction**: string - Optional style/emotion instruction, e.g. 'speak slowly and calmly'

## Examples

### Korean Preset Voice
Synthesize Korean speech with the Sohee preset
```json
{
  "model": "qwen/qwen3-tts",
  "input": {
    "mode": "custom_voice",
    "text": "안녕하세요, 코어닷투데이입니다. 무엇을 도와드릴까요?",
    "speaker": "Sohee",
    "language": "auto"
  }
}
```

### Voice Design from Description
Create a brand-new voice from a text description
```json
{
  "model": "qwen/qwen3-tts",
  "input": {
    "mode": "voice_design",
    "text": "Welcome aboard! This is your captain speaking.",
    "voice_description": "A deep, confident male voice with a calm authoritative tone"
  }
}
```

## Response Format
```json
{
  "job_id": "abc123",
  "status": "pending | processing | completed | failed",
  "result": "URL or data (when completed)"
}
```

## Usage Flow
1. POST /predictions with model and input -> receive job_id
2. GET /predictions/{job_id} -> poll until status is completed or failed
3. Result contains output URL(s)

## Tips
- Pick the mode first: preset voices (custom_voice), cloning (voice_clone), or description-based (voice_design)
- Sohee is the Korean preset speaker; language auto-detection handles mixed text
- In voice_clone mode, providing reference_text improves clone accuracy
- Use style_instruction like 'excited tone' to direct delivery without changing the voice

## Documentation
https://replicate.com/qwen/qwen3-tts