# MiniMax Speech 2.8 Turbo - Core.Today AI API

> Low-latency MiniMax Speech 2.8 Turbo with under 250ms latency, 40+ languages, voice cloning, natural interjections, and real-time pricing. Ideal for interactive and real-time applications.

- **Provider**: MiniMax
- **Model ID**: minimax/speech-2.8-turbo
- **Category**: Audio & TTS
- **Credits**: 142 per 1000 characters (0.135 credits/char, billed per character)
- **Speed**: Fast
- **Quality**: High

## Features
- Under 250ms latency for real-time use
- 40+ language support
- Natural interjections (laughs, sighs, coughs, etc.)
- Voice cloning from short audio samples
- Emotional expression control
- Pause markers and pronunciation control

## Use Cases
- Real-time voice assistants and agents
- Live streaming and interactive content
- Customer service chatbots
- Game character voices
- Multilingual content localization

## 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 narrate (max 10,000 characters). Use <#0.5#> to insert pauses. Supports interjections: (laughs), (sighs), (coughs), (gasps), (humming), (whistles), (sneezes), etc.

### Optional
- **voice_id**: string (default: English_Wiselady) - Voice preset or cloned voice ID. 17+ built-in voices available.
- **speed**: number (default: 1) - Speech speed multiplier (0.5-2.0)
- **volume**: number (default: 1) - Relative loudness. 1.0 is default MiniMax gain. Range 0–10.
- **pitch**: integer (default: 0) - Semitone offset (-12 to +12)
- **emotion**: string (default: auto) - Delivery style
  Options: auto, happy, sad, angry, fearful, disgusted, surprised, calm, fluent, neutral
- **english_normalization**: boolean (default: false) - Improve number/date reading for English text (adds a small amount of latency).
- **sample_rate**: integer (default: 32000) - Audio sample rate in Hz.
  Options: 8000, 16000, 22050, 24000, 32000, 44100
- **bitrate**: integer (default: 128000) - MP3 bitrate in bits per second. Only used when audio_format is mp3.
  Options: 32000, 64000, 128000, 256000
- **audio_format**: string (default: mp3) - File format for the generated audio. Choose mp3 for general use, wav/flac for lossless, or pcm for raw bytes.
  Options: mp3, wav, flac, pcm
- **channel**: string (default: mono) - mono for 1 channel (default), stereo for 2 channels.
  Options: mono, stereo
- **subtitle_enable**: boolean (default: false) - Reserved: the upstream currently returns audio only — no subtitle data is included even when enabled
- **language_boost**: string (default: None) - Language hint for better pronunciation
  Options: None, Automatic, Chinese, Chinese,Yue, Cantonese, English, Arabic, Russian, Spanish, French, Portuguese, German, Turkish, Dutch, Ukrainian, Vietnamese, Indonesian, Japanese, Italian, Korean, Thai, Polish, Romanian, Greek, Czech, Finnish, Hindi, Bulgarian, Danish, Hebrew, Malay, Persian, Slovak, Swedish, Croatian, Filipino, Hungarian, Norwegian, Slovenian, Catalan, Nynorsk, Tamil, Afrikaans

## Examples

### Interactive Voice Agent
Generate low-latency voice response with natural interjections
```json
{
  "model": "minimax/speech-2.8-turbo",
  "input": {
    "text": "안녕하세요! (laughs) 코어닷투데이에 오신 것을 환영합니다. <#0.3#> 최고의 AI 음성을 빠르게 경험해 보세요!",
    "voice_id": "Korean_CheerfulLittleSister",
    "emotion": "happy",
    "speed": 1,
    "language_boost": "Korean"
  }
}
```

## 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
- Best choice for real-time applications where latency under 250ms matters
- Use interjection tags like (laughs), (sighs) for more natural delivery
- Insert pauses with <#x#> markers for better pacing (0.01-99.99 seconds)
- Set speed to 1.1-1.2 for chatbot responses to feel more responsive
- Use language_boost for non-English text to improve pronunciation accuracy
- For production-quality output, consider Speech 2.8 HD instead
- subtitle_enable is accepted but the upstream does not return subtitle data yet — responses contain audio only

## Documentation
https://replicate.com/minimax/speech-2.8-turbo