Transcribe audio and video recordings into timestamped text transcripts.
npx skills add https://github.com/astralform-ai/skills --skill transcription-astralform-ai
Please help me install this Agent Skill. Skill: transcription Source: https://github.com/astralform-ai/skills/tree/main/skills/transcription Command: npx skills add https://github.com/astralform-ai/skills --skill transcription-astralform-ai
[MM:SS] lines so you can quote and link specific moments in a recording. - Flexible Input Sources: Accepts conversation asset ids, direct http(s) media URLs, or sandbox file paths, with automatic language detection or an ISO-639-1 hint. - Automatic Long-File Handling: Splits recordings over the ~25 MB provider limit and stitches parts back with absolute timestamps on the full timeline. - Use Case: A user attaches a 45-minute meeting recording and asks for a summary with key quotes. The Skill transcribes the whole file, then the assistant delivers the summary in the reply and the full transcript as an exported file. ## Quick Start Transcribe the attached meeting recording and give me a summary with the key quotes and their timestamps.High-intent search queries and answers about installing and using this skill.
Call transcribe_audio with the recording's asset id, direct URL, or sandbox path. It returns the transcript as [MM:SS] timestamped lines automatically, so you can quote and reference specific moments without requesting timestamps separately.
For YouTube, use the youtube skill first since published captions are faster than transcribing. For other pages, install yt-dlp with pip, extract audio with the -x flag to mp3, then pass the sandbox path to transcribe_audio.
Yes. The tool automatically splits files exceeding the ~25 MB provider limit and stitches the parts back with absolute timestamps on the full timeline. Do not pre-split or shorten the recording, and do not transcode formats like mp3, m4a, wav, ogg, opus, or flac.
No. Whisper, faster-whisper, and torch are already in the sandbox and the model weights ship with the image. Reinstalling them wastes time and disk space, and downloading weights will be slow or fail since model hosts are not reliably reachable.
A hosted provider normally finishes a 30-minute recording in under ten seconds. If it takes minutes, the tool fell back to in-sandbox transcription, which runs at roughly real-time speed but still produces a correct transcript.
jscraikExtracts transcripts, downloads media, and retrieves subtitles from video and audio sources using yt-dlp and ffmpeg.
openaiExtract verbatim customer quotes by theme from transcripts and meeting notes.
JigsawStackConvert spoken audio into searchable transcripts with timestamps and speaker labels.
bacchus-labsDownload YouTube transcripts and captions using yt-dlp.