DSH HUB
HomePlugin StorePlugin PacksCommunityRankingsResourcesPublish Guide
Plugin source
Back to catalog

chanchanxue /

chanchanxue/dsh-audio-extract

Verified

Lossless audio extraction from video for DeepSeek Harness

★ 0 Stars0 Forks0 IssuesN/A Community rating0 Confirmed installs
View on GitHub
READMESource: main@e1992620

dsh-audio-extract

Lossless audio extraction from video, as a DeepSeek Harness plugin.

It adds an 音频提取 / Audio extraction view to the session's view switcher, next to the built-in Chat and Trajectory views. Pick a folder, choose the videos, optionally trim the beginning and the end, fill in title/artist, pick a cover image and a lyrics file — the plugin stream-copies the audio track out of each video into an .m4a without re-encoding, then writes the metadata.

What it does

  • Lossless by construction — the audio is remuxed with -c:a copy; ffmpeg never decodes it, so the AAC frames stay byte-identical to the source.
  • Trims both ends — trimStart / trimEnd in seconds; the cut lands on AAC frame boundaries.
  • Metadata — title, artist, optional album and lyrics (©lyr), written into the MP4 container.
  • Cover art — a chosen image is embedded as an attached_pic stream (covr), so players and the iPhone Music app show it.
  • Folder picker + direct cover/lyrics selection — the folder comes from the native directory dialog; covers are uploaded through a local route and lyrics are read from a .lrc/.txt file.

Requirements

  • DeepSeek Harness with the Web UI (the plugin registers a conversation view and one session command).
  • ffmpeg available either on PATH, at C:\ffmpeg\bin\ffmpeg.exe, or pointed to by the DSH_AUDIO_EXTRACT_FFMPEG environment variable.
  • Source videos with AAC audio (mp4/mov/m4v/mkv/…); other codecs (AC-3, DTS) are rejected with a reason rather than silently re-encoded.

Install

From the plugin manager (Settings → Plugins), or:

dsh plugin --profile <profile> add github:chanchanxue/dsh-audio-extract

Use

  1. Open a session and pick 音频提取 from the view switcher at the top of the conversation.
  2. Choose the video folder — the videos are listed automatically.
  3. Set the start/end trim if needed, fill title/artist, optionally pick a cover and a lyrics file.
  4. Click 开始提取. The command result lists every produced file with size, duration and which metadata was written.

How it works

Piece Detail
View conversation.view, id audio-extract, order 31
Command /audio-extract <base64 JSON> — the page builds the payload and calls it through the session command API
Route GET /audio-extract/list?path=… Lists video files in a folder (the composed directory picker serves native and cannot browse)
Route POST /audio-extract/upload?name=… Stores an uploaded cover in the OS temp dir and returns its absolute path

Everything runs locally: no network calls, no uploads outside your machine, no telemetry.

Limitations

  • Output container is .m4a; it is only chosen for AAC sources. Non-AAC tracks need a different container and currently fail with a message.
  • Trimming is frame-accurate to the AAC frame (~23 ms), not sample-accurate — that is the trade-off for not re-encoding.
  • The cover/lyrics pickers use the browser's file input; the cover is copied into the OS temp folder.

License

MIT

—/ 5

No ratings yet

Verified DSH bundle

Commit e1992620d1b0

Community comments

No comments yet. Be the first to write one.

DSH HUB

A community index for DSH plugins. Not an official GitHub or DeepSeek AI product.

CommunityResourcesAPIAbout