Skip to content

Voice

Send a voice note and the bot transcribes it with Groq Whisper, shows you the transcript, and sends it to Claude as your message.

You need a Groq API key; Groq has a free tier. telecoder init asks for one, or add it to ~/.telecoder/.env:

Terminal window
GROQ_API_KEY=gsk_...
  • Language: English by default. Set VOICE_LANGUAGE (for example nl) for another.
  • Transcript: shown before Claude answers. VOICE_SHOW_TRANSCRIPT=false hides it.
  • Size: up to 19MB per note.

/transcribe transcribes without sending anything to Claude: send it as a reply to a voice note or audio file. Audio files and video aren’t sent to Claude on their own.

/tts turns on spoken replies for the chat or thread you’re in. Each reply then also arrives as a voice note, without its code blocks. The menu picks the voice, and whether it comes as a voice note that plays straight away or as an audio file.

Two providers:

  • Groq Orpheus, the default. It uses the same GROQ_API_KEY, and needs ffmpeg on the machine.
  • OpenAI: set TTS_PROVIDER=openai and OPENAI_API_KEY. TTS_MODEL, TTS_SPEED and TTS_INSTRUCTIONS tune it.

Replies longer than TTS_MAX_CHARS (4096) are cut at the end of a sentence. Every voice setting is in Configuration.