Skip to content
TextArray1.41.0
100% local

Voice transcript cleaner

Clean up raw speech-to-text transcripts by removing filler words, stutters and annotations.

Input
Output
Related tools
AI phrase removerText cleaning
Remove extra spacesText cleaning
Sentence splitterText manipulation
Remove lines starting with a numberText cleaning
TTS sanitizerText cleaning
Nickname cleanerText cleaning
Voice transcript cleaner
Paste a raw transcript exported from Whisper, Otter, Google's speech recognition or any other speech-to-text engine, and this tool turns the rough machine output into readable text. It strips hesitation sounds and filler words — "um", "uh", "erm", "hmm", "like", "you know", "I mean", "sort of" — along with the equivalents that turn up in Slovak, Czech, German, Polish and Hungarian dictation, so a mixed or non-English recording gets the same cleanup.
Speech recognizers also tend to transcribe stutters literally — "the the plan", "I- I- I will" — and drop bracketed markers where they lost the audio, like [inaudible], [music] or (laughs). Both are handled automatically: repeated words collapse to a single occurrence, and bracketed annotations disappear along with the extra spacing they leave behind. A final pass recapitalises the first letter of the text and the first letter after every period, question mark and exclamation mark, since transcripts often come back in a single run of lowercase words.
Each of the five cleanup steps — removing fillers, fixing capitalisation, collapsing repeats, dropping bracketed annotations, and merging separate lines into flowing paragraphs — can be switched on or off, so you can keep a transcript's original line breaks for subtitle work or merge everything into prose for an article or meeting summary. The tally under the output counts how many filler words and how many repeated words were removed, so you can see at a glance how rough the original recording was.
Everything runs locally in your browser. The transcript never leaves your device, which matters for meeting recordings, interviews or anything else that was not meant for a third-party server. Because the cleanup works on fixed word lists and patterns rather than genuine language understanding, always skim the result before publishing it — a legitimate "like" or "well" in the middle of a sentence can occasionally get swept up with the fillers.
FAQ
Which filler words does it remove?
The English set is "um", "uh", "erm", "hmm", "like", "you know", "I mean" and "sort of"/"kind of", plus common equivalents from Slovak, Czech, German, Polish and Hungarian dictation such as "ehm", "hm", "no", "teda", "proste", "ako", "halt", "also" and "hat". Turn the option off to leave every word exactly as transcribed.
How does it handle stutters like "I- I- I will go"?
An immediately repeated word — with or without a hyphen or comma between the repeats — is collapsed to a single occurrence, so "I- I- I will go" becomes "I will go" and "the the plan" becomes "the plan".
What counts as a bracketed annotation?
Anything inside square brackets or parentheses, such as [inaudible], [music] or (laughs) — the kind of marker a speech-to-text engine inserts when it could not transcribe a stretch of audio. Turning the option off keeps them in the output untouched.
What does "merge lines into paragraphs" do?
Speech-to-text exports often put every short segment on its own line. Enabling this joins the lines within each existing paragraph into flowing sentences, while a blank line in the original transcript still starts a new paragraph in the result.
Is my transcript uploaded anywhere?
No. Cleanup runs entirely in your browser using built-in word lists — the transcript is never sent to a server.