How does transcript detection handle Zoom, Teams, YouTube, and Otter?
Inspection scans the opening lines and cue patterns to classify teams, youtube, zoom, otter, srt, or text automatically before any cleaning runs. Timestamps, UUID cue IDs, voice tags, and colon-separated speaker prefixes normalize into uniform segments inside a browser Web Worker.
| Detected source | Identifying file & syntax markers | Speaker & timestamp parsing behavior |
|---|---|---|
teams |
<v Name> voice tags or UUID cue IDs |
Extracts speaker inside <v ...> and strips 8-4 hex UUID cue lines |
youtube |
<HH:MM:SS.mmm>/<c> tags or bare MM:SS lines |
Parses cue timings or standalone 0:12 timestamp headers |
zoom |
WEBVTT header or SRT cues with Name: prefix |
Splits speaker label before colon and preserves cue start stamp |
otter |
Speaker 0:14 or Speaker [00:14]: lines |
Captures speaker name, bracketed timestamp, and inline trailing text |
srt |
Arrow timing --> without voice tags |
Normalizes comma milliseconds into MM:SS or H:MM:SS labels |
text |
Unstructured prose without cue arrows | Reads non-empty lines and applies filler or inline cleanup cleanly |
Detecting voice spans, UUID cue IDs, and bare timestamps
Microsoft Teams .vtt downloads place a hexadecimal UUID identifier above every timestamp line and wrap each utterance inside <v Speaker Name>...</v> voice tags defined by the W3C WebVTT Voice Span and Timing Specification. Our inspector strips UUID noise automatically while lifting the voice attribute into an editable speaker label.
YouTube copy-paste dumps alternate bare 0:12 or 1:02:05 time lines with spoken sentences, whereas Otter.ai and Zoom plain exports place Speaker [00:14]: or Maya Lin 0:14 on the leading line. If you first need to shift cue timings or convert .ass or .sbv tracks before cleanup, use our timing editor to fix or convert raw SRT and WebVTT caption timing.
Stripping inline karaoke timing tags and HTML wrappers
Automated YouTube .vtt files embed inline word-level timestamps such as <00:00:01.200><c> inside every cue line, alongside occasional <b>, <i>, <u>, <font>, or <span> wrappers. Before deduplication runs, stripInlineTags strips those inline tags and collapses repeated whitespace into single spaces.
Cues without explicit <v> tags fall back to matching Speaker Name: utterance prefixes up to 36 characters long. If a cue contains zero spoken text after tag stripping, it is ignored cleanly so empty timing blocks never pollute the finished transcript.
Deduplicating rolling captions, speaker turns, and filler words
Raw video captions break spoken dialogue across short 3 seconds windows at 150 WPM and frequently repeat phrases verbatim across overlapping cues. Turning those fragments into readable meeting notes collapses rolling repeats 100% client-side, stitching fragmented clauses back into clean paragraphs aligned with the W3C WAI Audio and Video Transcripts Guide.
Trimming 3 to 16 word overlaps in rolling YouTube VTT
When Deduplicate rolling YouTube captions is checked, trimRollingOverlap compares each incoming line against the prior kept segment across four cases. It drops exact matches, ignores trailing substrings already present at the end of the previous cue, slices off cumulative prefixes when a cue repeats the previous sentence, and checks suffix-to-prefix word overlaps from 16 words down to 3 words.
Tip: Click Load sample VTT (
teams-architecture-review.vtt) to see Teams UUID cue removal, rolling sentence deduplication, speaker merging, andUm/Uhcleanup run locally in one pass.
Merging turns, renaming speakers, and stripping fillers
Keeping Merge consecutive same-speaker turns active joins adjacent lines by the same mapped speaker using smartJoin, attaching trailing punctuation directly without adding stray spaces before commas or periods. Dynamic Rename speakers inputs appear whenever speakers are detected, letting you replace Speaker 1 across every block without manual find-and-replace.
Checking Strip filler words (um, uh, er, ah, hmm) removes standalone vocal hesitations plus stretched variants while repairing orphaned double commas and leading sentence punctuation. Teams maintaining editable meeting minutes in LibreOffice Writer before distributing finalized packets can pair this workflow with our OpenDocument ODT to PDF converter.
Exporting cleaned notes to TXT, Markdown, DOCX, and PDF
Finished notes adapt to quick chat summaries, developer knowledge bases, executive Word reviews, or archival distribution depending on the output button you pick. Setting Document title (for MD / DOCX / PDF) adds a top-level document heading across structured downloads while keeping plain copy-paste clean.
Timestamp granularity across plain text and Markdown
Under Timestamps, Keep paragraph start time labels each merged speaker turn once with [MM:SS] in .txt or *(MM:SS)* in .md, while Remove all timestamps outputs pure dialogue prose. Selecting Keep every cue timestamp retains every individual cue boundary and disables adjacent turn merging so each timed line keeps its exact anchor.
Markdown exports (-clean.md) write # Title ATX headings, **Speaker** *(MM:SS)* attribution lines, and blank-line paragraph spacing matching the CommonMark 0.31.2 Markdown Specification. Live summary counts report detected source type, total blocks, word count, and removed rolling duplicates before download.
Local Word DOCX and paginated PDF report formatting
Clicking Export Word (.docx) builds a -clean.docx document directly from the cleaned Markdown using Calibri 11 pt typography on A4 pages. Clicking Export PDF compiles a -clean.pdf file with A4 portrait geometry, normal margins, 11 pt sans-serif text, the clean theme, and page numbers enabled.
All four file exports trigger browser downloads with -clean filename suffixes alongside a one-click Copy clean text button for Slack or Notion pasting. Writers preparing longer Markdown meeting reports for Word review can also export editable Word minutes without leaking private calls directly.
Keeping confidential meetings offline inside browser workers
Board reviews, employee feedback sessions, legal interviews, and production incident post-mortems contain sensitive personnel and strategic details that belong strictly offline without cloud uploads to third-party AI note services. Inspection, overlap deduplication, filler stripping, and DOCX packaging run inside the everyday Web Worker (runEverydayTask), while PDF layout runs inside the document Web Worker (runTask).
Word exports follow the Office Open XML package standard documented in the Library of Congress DOCX Transitional (ISO 29500) Profile, assembling document XML and styles locally in memory without external API calls. Inside Transcript Cleaner, every speaker rename edit and timestamp rule applies client-side, so no pasted meeting transcripts ever leave your browser tab.
After generating your .docx or .pdf file, ResultPanel displays the saved -clean filename, byte size, and elapsed conversion time locally. Once the page loads, text cleanup and export continue working even offline without remote queues. Security and IT responders documenting phishing escalations alongside meeting transcripts can trace raw SMTP routing headers locally using Email Header Analyzer.