Skip to main content
@mastra/voice-openai passes options to the OpenAI client, so a base URL override is enough.
Notes:
  • Audio is always WAV regardless of responseFormat, and transcription accepts WAV only: pass filetype: 'wav'.
  • The model names are typed as tts-1 / whisper-1 in the package; Lokutor ignores them.
  • language is forwarded in the request body (the package spreads extra options into it).
  • Not supported through this class: Mastra’s realtime voice (@mastra/voice-openai-realtime).
Verified against the live API with @mastra/voice-openai (speak, then listen on the result). A runnable copy is in lokutor-ai/openai-compatible-examples.