# Voice Transcribe

> Transcribes Telegram voice messages with a private whisper.cpp service on Ben's Latitude over Tailscale.

- Skill: `kip-claw/voice-transcribe` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add kip-claw/voice-transcribe`
- Raw SKILL.md: https://api.skillmd.com/api/skills/kip-claw/voice-transcribe/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: kip-claw (https://skillmd.com/u/kip-claw)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/kip-claw/voice-transcribe

---


# Voice Transcription

Telegram voice messages are transcribed by a token-authenticated `whisper.cpp` service on Ben's Latitude. The Pi converts audio transiently and sends it only over Tailscale; it has no local Whisper fallback.

When a voice message arrives:

1. Download the OGG file from Telegram to a temp path.
2. Run: `{{HOME}}/bin/voice-transcription-runner.sh <path_to_ogg>`.
3. Use the stdout output as the user's text input.
4. Delete the temp OGG file after transcription.

The transcript should be treated as a normal text message from the user.
If transcription fails or produces no output, reply: "I couldn't make
out that voice message — could you try again or type it out?"