Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.burki.dev/llms.txt

Use this file to discover all available pages before exploring further.

OpenAI TTS

OpenAI TTS is supported in Burki through the openai TTS provider. Use it when you already use OpenAI keys or want GPT-4o mini TTS voice instructions.

​
Quick Setup

1

Get API Key

Create an OpenAI API key from the OpenAI dashboard.
2

Configure in Burki

Select openai as the TTS provider and add the API key at the assistant or organization level.
3

Choose Model and Voice

Pick a supported model_id, voice_id, and optional speed/instructions settings.

​
Configuration

{
  "tts_settings": {
    "provider": "openai",
    "voice_id": "alloy",
    "model_id": "gpt-4o-mini-tts",
    "provider_config": {
      "speed": 1.0,
      "instructions": "Speak warmly and clearly, like a helpful support agent."
    }
  }
}
OptionTypeDescription
providerstringUse openai
voice_idstringOpenAI voice ID
model_idstringtts-1, tts-1-hd, gpt-4o-mini-tts, or pinned GPT-4o mini TTS snapshot
speednumberSpeaking speed, clamped between 0.25 and 4.0
instructionsstringVoice style instructions for gpt-4o-mini-tts* models

​
Supported Models

Model IDNotes
tts-1Standard OpenAI TTS model optimized for speed
tts-1-hdHigher quality OpenAI TTS model
gpt-4o-mini-ttsGPT-4o mini TTS model with instruction support
gpt-4o-mini-tts-2025-12-15Pinned GPT-4o mini TTS snapshot configured in the backend

​
Supported Voices

Voice IDDescription
alloyBalanced and clear
echoDeep and resonant
fableWarm and expressive
onyxStrong and authoritative
novaBright and energetic
shimmerSoft and gentle
ashNatural and conversational
balladMelodic and expressive
coralWarm and friendly
sageCalm and wise
verseSmooth and articulate
marinClear and professional
cedarWarm and natural

​
Runtime Behavior

  • Burki requests PCM audio from OpenAI and converts it for the current telephony path.
  • gpt-4o-mini-tts* models support the instructions field.
  • HD telephony providers use linear16 at 16 kHz where supported; Twilio-style paths use µ-law at 8 kHz.
  • Sentence buffering is used so the first audio chunk can start quickly without waiting for the entire LLM response.

​
When to Use OpenAI TTS

  • You want a simple provider stack with OpenAI for both LLM and TTS.
  • You need model-level voice instructions.
  • You prefer built-in voices over custom voice cloning.
For broader multilingual voice cloning, compare ElevenLabs, Cartesia, and Inworld.