Convert VTT to CSV Online

How to Convert VTT to CSV Online

1. Upload Your VTT File: Drag and drop your .vtt file or click "Add Files" to browse. Parsing happens in your browser — nothing leaves your device. Batch conversion is supported, so you can drop a multi-episode caption bundle and download one CSV per file (or stitch them after).
2. Confirm Output Format: CSV is preselected as the target. The converter walks every WebVTT cue, extracts the start timestamp, end timestamp, cue identifier (when present), <v Speaker> voice tag (when present), and cue text — then writes one row per cue with a header line for the spreadsheet.
3. Review the Preview (Optional): WEBVTT header, NOTE comments, STYLE blocks, and REGION blocks are dropped because they don't map to tabular data. Cue settings such as line:, position:, align:, vertical:, and size: are also stripped. Inline tags like <b>, <i>, <u>, <c.classname>, and <ruby> are removed from the text column so the spreadsheet reads as plain prose.
4. Convert and Download: Click Convert. The .csv file downloads immediately as UTF-8 with comma separators and RFC 4180 quoting, ready to open in Excel, Google Sheets, Numbers, or any pandas/R/SQL import pipeline.

Why Convert VTT to CSV?

WebVTT is a W3C-standardised plain-text format built for the HTML5 <track> element — its cues, settings, and STYLE blocks make sense to a browser but not to a spreadsheet. CSV flattens each cue into a row of plain values: index, start, end, speaker, text. Once your captions live in tabular form, every analytics, translation, and search tool that already speaks CSV can ingest them without a custom parser.

• Translation and localisation — Translators prefer CSV because each cue lands on its own row with source text in one column and an empty target column for the translation. memoQ, Trados, Smartcat, and Crowdin all import CSV cleanly; round-tripping back to VTT after translation preserves timings exactly.
• Word-count and reading-speed audits — Broadcasters and streamers enforce reading-speed limits (commonly 17 characters per second for adult content, 12 cps for children's programming per BBC Subtitle Guidelines). Drop the CSV into Excel, compute LEN(text) / (end_seconds - start_seconds), and flag any row over your threshold.
• Searchable transcripts in spreadsheets — A two-hour lecture's WebVTT may contain 1,500+ cues. Once in CSV, Ctrl+F or filter-by-keyword finds every occurrence of a term across an entire course in seconds, with timestamps to jump back to the video.
• Database ingestion for transcript search — Postgres COPY, MySQL LOAD DATA INFILE, BigQuery, and Snowflake all consume CSV natively. Loading a year of podcast transcripts into a searchable database starts with a VTT-to-CSV pass.
• AI / ML training data prep — pandas, datasets, and Hugging Face loaders read CSV in one line. Researchers building speech-to-text fine-tuning sets pair audio segments with cue text via the start/end columns.
• QA and proofreading — Editors hate scrolling through a .vtt file looking for typos. In a spreadsheet they can sort by length, filter by speaker, freeze the timestamp column, and tab through rows with the keyboard.

Need a different starting format? Try SRT to CSV if your captions came from a desktop editor, TTML to CSV for broadcast deliveries, or ASS to CSV for anime/fansub sources. Reverse the direction with CSV to VTT once you've finished editing.

VTT vs CSV — Format Comparison

Property	VTT (WebVTT)	CSV
Standardised by	W3C webvtt1 Candidate Recommendation Draft	RFC 4180 (informational)
File extension	.vtt	.csv
MIME type	text/vtt	text/csv
Structure	Cue blocks separated by blank lines	One row per record, comma-separated
Timestamps	Required, hh:mm:ss.ttt or mm:ss.ttt	Free-form column (kept as text by default)
Header line	Required: WEBVTT	First row is column names (by convention)
Inline tags	<b> <i> <u> <c> <v> <ruby> <lang>	None — plain text only
Cue settings	line: position: align: vertical: size:	Not representable
STYLE / REGION / NOTE blocks	Yes — embedded CSS, regions, comments	Not representable
Reading-speed analysis	Requires custom parser	One formula in any spreadsheet
Tooling	HTML5 video players, captioning editors	Excel, Google Sheets, pandas, SQL, R, jq
Encoding	UTF-8 (mandated by spec)	UTF-8 recommended; legacy tools sometimes default to CP1252

Output CSV Columns — What Each Cue Becomes

The output is a standard comma-separated file with a header row. One WebVTT cue becomes one row:

Column	Source in the VTT cue	Example value
index	1-based sequential counter	1, 2, 3
identifier	Optional cue identifier line (the line before the timestamp)	cue-12, intro, or blank if absent
start	Start timestamp before the -->	00:00:22.230
end	End timestamp after the -->	00:00:24.606
speaker	Content of <v Speaker>...</v> voice tag, if present	Alice, blank if none
text	Cue payload with inline tags stripped, line breaks joined with a space	Welcome to the show.

The first row is the header (index,identifier,start,end,speaker,text). Values containing commas, double quotes, or line breaks are wrapped in double quotes per RFC 4180; embedded quotes are doubled (""). The file is written as UTF-8 without a BOM, which is what Google Sheets, modern Excel (Microsoft 365 and Excel 2019+), Numbers, and pandas all expect.

Frequently Asked Questions

What CSV columns does the output file contain?

By default: index, identifier, start, end, speaker, text. The index column is a 1-based counter the converter adds (WebVTT cue numbering is optional in the spec). The identifier column carries any non-numeric cue ID from your source file and is blank when the cue had none. The speaker column is populated only when a cue uses a <v Speaker> voice tag — otherwise it's empty. If you need only start, end, text, delete the unwanted columns in Excel after export; the rows themselves are untouched.

Why are my inline tags like <b> and <i> missing in the CSV?

Because CSV is plain text — there's no way to render bold or italics in a spreadsheet cell. The converter strips <b>, <i>, <u>, <c.class>, <ruby>, <rt>, <lang>, and timestamp tags from the text column so the prose reads cleanly when imported. Plain characters (including emoji and accented characters) are preserved exactly. If you need to keep the markup, convert to VTT to SRT or VTT to TXT instead — SRT preserves the supported tags verbatim.

How does the converter handle WebVTT NOTE, STYLE, and REGION blocks?

They're discarded. NOTE blocks are author comments not intended for display, STYLE blocks contain CSS that has no meaning in a spreadsheet, and REGION blocks define on-screen positioning that CSV can't represent. The converter walks the file per the W3C parsing rules, recognises these block types, and skips them — only actual cues become rows.

What format are the timestamps in the start and end columns?

Exactly as they appear in the source VTT: hh:mm:ss.ttt when the cue has hours (e.g., 01:23:45.678) or mm:ss.ttt when it doesn't (e.g., 02:15.500). Excel and Google Sheets both treat these as text strings, which is what you want — leave them alone if you're just searching or editing, or convert to seconds in a formula (=TIMEVALUE(...)*86400) when you need numeric math for reading-speed audits.

Will commas, quotes, and line breaks inside a cue break my spreadsheet?

No. The converter follows RFC 4180 quoting: any field containing a comma, a double quote, or a CR/LF gets wrapped in double quotes, and any embedded double quote is escaped as "". Modern Excel, Google Sheets, LibreOffice Calc, Numbers, and the Python csv module all parse this correctly. If you're using a legacy importer that chokes on quoted fields, open the CSV in a text editor and you'll see the quoting is well-formed.

My CSV looks garbled in Excel — what's wrong?

Almost always an encoding mismatch. The converter writes UTF-8 without a BOM. Excel on Windows still defaults to CP1252 when you double-click a .csv, so accented characters and CJK glyphs render as mojibake. Two fixes: (1) in Excel, use Data → From Text/CSV and explicitly pick 65001: Unicode (UTF-8) in the dialog; or (2) save a copy of the CSV with a UTF-8 BOM (EF BB BF at the start) and Excel will autodetect it. Google Sheets and Numbers both autodetect UTF-8 without help.

Can I get a row per word or per sentence instead of per cue?

The default is one row per cue, which mirrors the WebVTT source 1:1. For per-word output you'd need WebVTT timestamp tags inside cues (<00:01:23.500>word) — those are stripped during conversion. If your VTT was generated by a speech-to-text engine that emits word-level cues (each word is its own cue), the CSV will already have one word per row. For per-sentence output, run the conversion and then split rows on sentence boundaries in your spreadsheet or pandas DataFrame.

Does the speaker column populate from NOTE Speaker: lines too?

No — only from <v Speaker>...</v> voice tags inside cues, which is the W3C-defined mechanism for marking who's talking. Many transcription tools embed speaker names inline in the text instead (Alice: Welcome back.) — in that case the speaker column stays blank and the name lives in the text column. You can split it out with =LEFT(F2, FIND(":", F2)-1) in Excel or df['text'].str.split(':', n=1, expand=True) in pandas.

Can I convert back to VTT after editing in a spreadsheet?

Yes — that's what CSV to VTT is for. Edit the text column, save as CSV (UTF-8), and the reverse converter rebuilds a valid WebVTT file with the WEBVTT header, the original timestamps, and any speaker tags wrapped back into <v> syntax. This is the standard translation round-trip: VTT → CSV → translator works in Excel → CSV → VTT.

Is anything uploaded to your servers?

No. The VTT is parsed and the CSV is generated entirely in your browser session — there's no upload, no account, no watermark, and no rate limit. The file is deleted when you close the tab.