QwenASRMiniTool CLI
QwenASRMiniTool turns audio and video into SRT subtitles on your own computer, and Claude Code drives it via the qwenasr skill.
- 578 ★ on the repository
- Last commit
Extended access
Why this level
- The agent runs a local executable, and the built-in API and Cloudflare tunnel expose access externally when enabled.
Install
- Works in
Manual install
mkdir -p ~/.claude/skills/qwenasr
cp -r .claude/skills/qwenasr/. ~/.claude/skills/qwenasr/The README says to copy `.claude/skills/qwenasr/SKILL.md` to `~/.claude/skills/qwenasr/`. The program itself comes from the repository releases.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent runs a local executable, and the built-in API and Cloudflare tunnel expose access externally when enabled.
This is third-party code. Review the repository files before installing.
What it does
QwenASRMiniTool is a Windows program with a WebView2 window that recognizes speech locally with Qwen3-ASR and Whisper models: audio, video and microphone recordings become SRT subtitles. Data stays on the computer. It runs on CPU through OpenVINO INT8 and on GPU through CrispASR. With a subcommand the program runs windowless as a CLI: `transcribe`, `status`, `profiles`, `apply`. The `apply` command writes only corrected text back into the original timings, so the agent cannot break timestamps. The `.claude/skills/qwenasr/SKILL.md` skill teaches Claude Code to call the CLI, pick a model profile and fix recognition errors.
Who it is for. Editors, journalists and students who need transcripts and subtitles without uploading recordings to the cloud, with text cleanup through Claude Code.
Good fit when
- An interview recording becomes a draft transcript on a laptop without internet.
- A video gets SRT subtitles together with speaker labels.
- An agent fixes names and terms in a transcript without moving timings.
- Japanese anime lines and songs are transcribed by a specialized model.
- A phone or another device sends a file through the built-in OpenAI-compatible API.
Not a fit when
- Mac or Linux is required: the README build targets Windows and WebView2.
- Live broadcast needs continuous word output: audio is split by pauses rather than streamed.
- Heavy background music plays and the `tea` profile is considered: on such material it may return nothing.
Example request
Transcribe interview.mp3 with qwenasr using the participant names as a hint, fix only the text and save the result to final.srt with apply.Limitations
The program targets Windows, and the executable paths in SKILL.md are Windows paths that you will need to adjust. On first run models download at 1 to 2.5 GB, which takes several minutes. The repository lists no license, so confirm usage terms with the author. The interface and docs are in Chinese, mostly in traditional script. The `twmix` profile is fast but outputs no punctuation. The last commit dates from September 2026.
How to disable. Delete `~/.claude/skills/qwenasr/` and the program itself along with downloaded models. Disable the API endpoint and the Cloudflare tunnel in the program settings.
FAQ
How do I get a transcript from the command line?
Run `QwenASR-WebView.exe transcribe audio.mp3 --json`. Stdout returns JSON with segments and times, and the SRT file is saved to the `subtitles` folder.
Why is apply needed?
The agent edits only the `text` field in JSON, and `apply fixed.json --base work.json` takes timings from the original file. Timestamps stay intact.
Which model profiles exist?
`zh` for ordinary speech, `tea` for clean Taiwanese Mandarin, `ja` for Japanese and anime, `twmix` for mixed Mandarin and Taiwanese.
Do recordings go to the cloud?
Recognition runs locally. The Cloudflare tunnel for external access is enabled separately, and files pass through it when it is used.
Which formats are accepted?
Audio MP3, WAV, FLAC, M4A, OGG and video MP4, MKV, MOV: ffmpeg extracts the sound from video.
Discussion
No questions or reviews yet.
Related
CLI toolsVideoCaptioner transcribes speech, translates subtitles and burns them into video with a single command from Claude Code.
The eyes CLI and skill for agents: a video index, search by words and on-screen text, close-up frames and editing metrics
A skill that transcribes video, lets you review captions in a browser and burns in Chinese and English captions.
Claude Real Video
claude-real-video
A skill that lets Claude watch a video: it pulls scene-aware, deduplicated frames and a speech transcript through whisper from a link or file
The skill turns a video or podcast link or file into a transcript and summary using platform subtitles and Whisper.
A Codex and Claude Code skill: subtitles and a summary from a Douyin, Xiaohongshu, Bilibili or YouTube link or a local file
README in short
The project aim is free offline speech recognition on anyone's own computer. Since version 2.0.0 only the WebView edition is maintained.