telegram-voice-transcription

name	telegram-voice-transcription
description	Transcribe saved Telegram voice notes and other audio attachments with OpenAI before replying. Use when a Telegram prompt includes an audio file path under inbox/telegram/.
compatibility	Requires OPENAI_API_KEY or HERMIT_OPENAI_API_KEY, network access to api.openai.com, and a saved local audio file.

Telegram Voice Transcription

Use this skill when a Telegram message includes a saved audio attachment such as:

• inbox/telegram/...voice-note.ogg
• inbox/telegram/...audio.mp3
• inbox/telegram/...video-note.mp4

What it does

• sends the local audio file to OpenAI transcription
• defaults to whisper-1
• writes a sidecar transcript file next to the audio
• prints the transcript to stdout so you can use it immediately

Usage

From the repo root:

node skills/telegram-voice-transcription/transcribe.mjs workspace/inbox/telegram/<file>

Examples:

node skills/telegram-voice-transcription/transcribe.mjs workspace/inbox/telegram/00000011-000013-voice-note.ogg

node skills/telegram-voice-transcription/transcribe.mjs workspace/inbox/telegram/00000011-000013-voice-note.ogg --model whisper-1

node skills/telegram-voice-transcription/transcribe.mjs workspace/inbox/telegram/00000011-000013-voice-note.ogg --stdout-only

Output

By default the script writes:

• <audio-file>.transcript.md

The file contains:

• source audio path
• model used
• timestamp
• transcript text

Recommended workflow

1. Run the script on the saved Telegram audio file.
2. Read the transcript.
3. Answer the human based on the transcript.
4. Use send_telegram_message or send_telegram_voice_note for the reply.

Notes

• If the audio is unclear, say that plainly instead of pretending it was clear.
• If transcription fails because the API key or network is missing, tell the human and ask for text instead.