Transcribe Video
Turn a recording into editable text, subtitles and word timings using your computer’s GPU or CPU.
Open your local studio
- On your Windows computer, open the repo’s
local-transcriberfolder and double-click Start Transcribe Video.cmd. - It runs in the background, with no window to keep open. Open the studio below; its Stop button at the bottom ends it.
- Choose a video, transcribe it, review the text and download your exports.
If the page cannot connect, start the launcher first. The address opens a service on the device you are using. Your phone cannot use the engine on your PC.
The workflow
From a long talking video to posted short-form clips.
- 01 Record Film one long talking video on camera. You
- 02 Transcribe Run the video through this tool to get timed text and word JSON. This tool
- 03 Copy the brief Use Full edit plan + content package in the local studio, then set your audience and goals. This tool
- 04 Get the plan Paste the brief into ChatGPT or Claude for clip picks, source ranges and captions. ChatGPT or Claude
- 05 Edit in CapCut Claude drives CapCut on your computer to cut the clips, set the framing (9:16 for Shorts, Reels and TikTok, or 16:9 for YouTube) and export. Claude + CapCut
- 06 Review and post Check each clip, then add the captions and hashtags before you post. You
Steps 02 and 03 happen on your computer. Recordings stay local.
Text, captions and word timings
Choose Clean voice first to remove background noise before transcribing: one upload gives you the cleaned video and the transcript together. It needs Clean Voice installed on the same computer.
Edit transcript text and sentence timings, jump to a passage in your recording, and export TXT, SRT, VTT, word-timing JSON and review notes.
Use Full edit plan + content package in the local studio to copy a detailed AI brief with your word JSON. Customize the audience, platform and video goals, then paste it into ChatGPT or Claude for suggested edits, clips and publishing copy.
Recordings stay on your computer. First use downloads the selected speech model; cached models can run offline. No paid transcription API is used.