MotionShine01 / Find
Every word, timed
An AI speech model writes out each word with when it starts and ends, the ums and uhs too.
Free tool
Remove um and uh, keep what you meant.
Drop in a voice-over, a podcast or a screen recording with your voice, and an AI speech model finds every "um", "uh" and "hmm", plus the long pauses where you stopped to think. Check the list, listen to any cut, untick what should stay, and get a tighter recording or video back. It all runs in this tab, so nothing is uploaded.
Transcript with the cuts struck through
What to cut
Shortened pauses keep a gap of up to 0.4 seconds, so the speech still breathes. A video is written again as an MP4.
Your recording is processed in your browser. Nothing is uploaded.
How it works
Find, check, cut.
02 / Check
A list of cuts
Each filler and long pause is listed with its time. Press Hear to listen around it, untick any you want to keep.
03 / Save
A cleaner file
The cuts are joined with short fades so they do not click. Video comes back as MP4, sound as MP3 or WAV.
Questions
Good to know.
Is my recording uploaded?
No. The speech model runs in this tab, and the new file is made here too. The first time, the page downloads the model from Hugging Face, about 80 to 95 MB, and keeps it for next time.
Which model finds the fillers?
Whisper base from OpenAI (Apache 2.0 licence), in a build by onnx-community that gives the time of every word. Unlike many transcribers, this build keeps words like "um" and "uh", so they can be found.
Will the cuts be noticeable?
Each cut is joined with a 10 millisecond fade and a short gap is kept, so speech still sounds natural. In a video, the picture jumps at each cut, as with any jump cut; screen recordings with little movement hide this best.
Which languages?
English. The filler list is for English speech, such as um, uh, er, ah and hmm.
Does it change the video quality?
The video is encoded again as H.264 at a high bitrate, so it looks the same for most uses. The original file is not changed.