OpenAI Text-to-Speech (TTS)

Overview

Generate voice notes using OpenAI's TTS API. Useful for sending audio responses to users on Telegram, WhatsApp, etc.

Requirements

• OPENAI_API_KEY environment variable set

Quick Usage

curl -s https://api.openai.com/v1/audio/speech \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini-tts",
    "input": "Your text here",
    "voice": "onyx"
  }' \
  --output /tmp/voice.mp3

Then send with: MEDIA:/tmp/voice.mp3

Available Voices

Models

Recommended: Use gpt-4o-mini-tts — better quality and 25x cheaper than legacy models.

Parameters

{
  "model": "gpt-4o-mini-tts",
  "input": "Text to speak",   // Max 4096 characters
  "voice": "onyx",            // See voices above
  "response_format": "mp3",   // mp3, opus, aac, flac, wav, pcm
  "speed": 1.0                // 0.25 to 4.0
}

Examples

Norwegian voice note

curl -s https://api.openai.com/v1/audio/speech \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini-tts",
    "input": "Hei! Dette er en stemmenote på norsk.",
    "voice": "nova"
  }' \
  --output /tmp/norwegian.mp3

With slower speed

curl -s https://api.openai.com/v1/audio/speech \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini-tts",
    "input": "Important announcement...",
    "voice": "onyx",
    "speed": 0.9
  }' \
  --output /tmp/announcement.mp3

Opus format (smaller files)

curl -s https://api.openai.com/v1/audio/speech \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini-tts",
    "input": "Compact audio message",
    "voice": "alloy",
    "response_format": "opus"
  }' \
  --output /tmp/message.opus

Workflow

1. Generate audio with curl
2. Save to /tmp/voice-{timestamp}.mp3
3. Send with MEDIA:/tmp/voice-{timestamp}.mp3

Best Practices

• Use onyx or echo for professional/calm tone
• Use nova for friendly/upbeat messages
• Keep messages concise (under 1 minute ideal)
• Use opus format for smaller file sizes
• Always use gpt-4o-mini-tts (best quality, lowest cost)