Groq Voice Transcribe

Fast speech-to-text for voice notes and audio files through Groq's OpenAI-compatible transcription endpoint.

Use it when you want cloud transcription via Groq instead of running Whisper locally.

Best for:

• Telegram / Signal voice notes
• short audio clips
• Chinese, English, or mixed daily speech
• fast transcript generation for follow-up summarization or chat replies

What you need

You need a Groq API key.

Groq often provides a free developer tier / trial credits for new users.

Get one from:

• https://console.groq.com/
• Sign in, open API Keys, then create a key

Easiest setup in OpenClaw

If OpenClaw is already running and configured, you can simply ask your assistant:

• "Configure Groq Voice Transcribe for me"
• "Here is my Groq API key, set up Groq Voice Transcribe"

The assistant can place the key into ~/.openclaw/openclaw.json for you.

Manual setup

Set GROQ_API_KEY, or configure it in ~/.openclaw/openclaw.json under:

{
  "skills": {
    "entries": {
      "groq-voice-transcribe": {
        "apiKey": "GROQ_KEY_HERE"
      }
    }
  }
}

Quick start

{baseDir}/scripts/transcribe.sh /path/to/audio.ogg

Defaults:

• Model: whisper-large-v3-turbo
• Output: .txt
• Format: plain text

Common examples

# Basic transcript
{baseDir}/scripts/transcribe.sh /path/to/audio.ogg

# Chinese voice message
{baseDir}/scripts/transcribe.sh /path/to/audio.ogg --language zh --prompt "中文普通话，日常聊天"

# Save to a custom file
{baseDir}/scripts/transcribe.sh /path/to/audio.ogg --out /tmp/transcript.txt

# Verbose JSON output
{baseDir}/scripts/transcribe.sh /path/to/audio.ogg --json --out /tmp/transcript.json

Flags

• --model : transcription model (default whisper-large-v3-turbo)
• --out : output file path
• --language : hint the spoken language, for example zh, en, ja
• --prompt : optional context or spelling hint
• --json: write verbose JSON instead of plain text

Notes

• Audio is sent to Groq for transcription.
• This skill is meant for transcription, not text-to-speech.
• If language is omitted, Groq can usually auto-detect it, but passing --language zh often helps for Chinese voice notes.