| name | audio-transcription |
| description | Transcribe audio and video files into structured notes. Activate this skill when users want to transcribe recordings, meetings, podcasts, voice memos, or any audio/video content in their vault. |
Audio & Video Transcription
Transcribe audio and video files from the vault into structured Obsidian notes using read_file to send binary media directly to the model.
Supported Formats
Audio: .wav, .mp3, .aac, .flac, .webm (audio-only)
Video: .mp4, .mpeg, .mov, .flv, .mpg, .webm, .wmv, .3gp
Size limit: 20 MB per file (Gemini inline data limit).
How to Transcribe
- Use
read_file with the path to the audio/video file — the binary data is sent directly to the model for processing.
- Listen to/watch the content and produce a transcription.
- Use
write_file to save the transcription as a markdown note.
Transcription Format
Structure transcriptions as follows:
---
tags:
- transcription
source: "[[original-file.mp3]]"
date: YYYY-MM-DD
duration: "MM:SS" (estimate if possible)
---
# Transcription: [Title]
## Summary
Brief 2-3 sentence summary of the content.
## Transcript
[00:00] Speaker 1: Opening remarks...
[00:45] Speaker 2: Response...
[01:30] Speaker 1: Follow-up...
Guidelines
- Timestamps: Include approximate timestamps in
[MM:SS] format at natural breaks (new speakers, topic changes, pauses).
- Speaker identification: Label distinct speakers as "Speaker 1", "Speaker 2", etc. If names are mentioned, use them after first identification.
- Filler words: Omit excessive filler words (um, uh, like) unless they carry meaning.
- Inaudible sections: Mark unclear audio as
[inaudible] or [unclear].
- Non-speech sounds: Note significant sounds like
[laughter], [applause], [music].
- Summary: Always include a brief summary at the top for quick reference.
- Frontmatter: Link back to the source file using a wikilink.
Tips
- For long recordings, let the user know the transcription may be partial due to the 20 MB size limit. Suggest splitting large files with an external tool.
- If the user asks to "transcribe the recording in this note", use
read_file on the current note first to find embedded audio/video links (e.g., ![[recording.mp3]]), then read_file on the linked file.
- For meeting notes, suggest adding attendees and action items sections after the transcript.
- For podcasts or interviews, suggest adding a "Key Topics" section with timestamps.
