Transcript Cleaner: SRT, VTT and Zoom to Clean Text
Clean SRT, VTT, Zoom, Otter, and YouTube transcripts: remove timestamps and duplicates, merge speaker turns, and split into LLM context window chunks.
Input Transcript
0 lines • 0 words • 0 charsCleaned Output
0 words 0 chars 0 est. tokens 0 speakers 00:00 audio est.Cleaned transcript will appear here instantly...
Speaker Talk-Time & Breakdown
Turn volume and audio speaking duration analysis| Detected Speaker | Rename In Output | Talk Time Share | Est. Duration |
|---|---|---|---|
| No speakers detected | |||
Why Clean Transcripts Before Sending to AI Models?
Subtitle and caption files from Zoom recordings, Microsoft Teams, Otter.ai, Google Meet, YouTube, and Whisper contain heavy structural noise. Pasting raw subtitle data directly into ChatGPT, Claude 3.5 Sonnet, or Google Gemini wastes massive amounts of token budget and degrades summary accuracy:
Timestamp Overhead
Numeric timestamp cues like 00:01:23,400 --> 00:01:28,900 consume up to 40% of prompt context without adding semantic value. Removing timestamps yields dramatically longer prompt capacity and cheaper API calls.
YouTube Rolling Duplicates
Live stream and YouTube auto-generated captions refresh in sliding 3-line windows. Words are repeated across two to three consecutive cues, creating garbled text unless reconciled with overlapping n-gram deduplication.
Mid-Sentence Cue Breaks
Speech-to-text engines break sentences mid-phrase to fit video screen widths. Merging turns by the same speaker reconstructs natural syntax so AI attention mechanisms can analyze coherent arguments.
Supported Formats & Standards
This cleaner automatically detects and processes every common subtitle and transcript standard:
- SubRip (.SRT): Numeric cue headers and comma-separated millisecond timestamps.
- WebVTT (.VTT): Period timestamps, cue identifier headers, and voice tags like
<v Speaker Name>. - Zoom & Microsoft Teams: Time-stamped attendee dialog logs (
00:02:14 John Doe: ...). - Otter.ai & Descript: Multi-line speaker paragraphs with audio time markers.
- YouTube Transcripts: Copied YouTube transcript sidebars with seconds markers and rolling phrase repetitions.
- JSON Speech Segments: Array of objects containing
{ "speaker": "...", "text": "..." }.
Frequently Asked Questions
How does the context window chunker work?
The chunker groups consecutive speaker segments until the estimated token limit (e.g. 2,500 tokens) is reached, without severing mid-sentence thoughts. Each chunk is packaged with customizable summary prompts and can be exported as individual text files or a single ZIP archive.
How are YouTube rolling duplicates eliminated?
Our sliding n-gram suffix-prefix reconciliation algorithm compares the trailing words of previous cues against the leading words of upcoming cues. Overlapping duplicate words are automatically trimmed to produce clean, natural prose.
Is my confidential meeting audio transcript private?
Yes. The entire application executes strictly inside your local browser via vanilla JavaScript and Web APIs. No transcript text, speaker names, or credentials ever leave your machine.