Video to Transcript — Any File or URL

Key takeaways

• ●File upload or URL paste — both work. Nine video file formats (MP4, MOV, WebM, MKV, AVI, M4V, MPEG, FLV, WMV) plus five URL platforms (TikTok, Instagram, Vimeo, Loom, direct video links). Audio is extracted server-side; no ffmpeg on your end.
• ●Free preview before signup. The first 200 characters of any transcript render immediately with the same timestamps and speaker labels the full output uses. See exactly what you'll get before creating an account.
• ●20–40 seconds for a typical 3-minute video. A 30-minute podcast video finishes in 4–7 minutes. AI runs at 4–10× real-time depending on load.
• ●95% accuracy on clean audio, 80–95% on real-world video. Studio and podcast content lands at the top of that range; vlogs and TikTok with music underneath speech land at the bottom.
• ●Language auto-detects from your browser. Turkish browser → Turkish pre-selected, Spanish browser → Spanish, and so on. Fixes the "why is my English video's transcript in Spanish" bug that happens when tools blindly pick the first available caption track.
• ●Speaker labels on every transcript. Automatic diarization for up to eight distinct speakers. Rename in the editor and the change applies across the whole transcript.
• ●YouTube goes to a different tool. /tools/youtube-transcript is faster and cheaper for YouTube specifically because we pull YouTube's caption track when it exists. This page is for everything else.

How it works (3 steps)

The workflow is the same for a 15-second TikTok and a 3-hour lecture recording. Paste or drop, preview, sign up to unlock.

1. 1

   Paste a URL or drop a file

   Drop any supported video file into the uploader, or paste a public URL from TikTok, Instagram, Vimeo, Loom, or any direct video link. Language auto-detects from your browser; override in the dropdown if the video is in a different language than your locale.

2. 2

   Preview the first 200 characters

   The server fetches the video (URL) or reads the file (upload), extracts the audio track, and runs Whisper Large-v3. In 20–40 seconds you see the first 200 characters of the transcript with timestamps and speaker labels — enough to confirm the language, quality, and speaker structure are what you expect.

3. 3

   Sign up to unlock and export

   A free account (email only, no credit card) reveals the full transcript, the editor with search-and-replace, and export to TXT, DOCX, SRT, VTT, and JSON. Translate to any of 133 languages from the same editor. Free tier includes 30 minutes/month; paid from $2/month.

Supported formats and URLs

Two ways to hand VexaScribe a video: upload a file, or paste a URL. Anything on the two lists below works without pre-processing.

File formats

Video URLs

Not supported by this tool

Audio-only files also work: MP3, M4A, WAV, AAC, FLAC, OGG, OPUS. For dedicated audio guides see MP3 to text, M4A to text, WAV to text, and OGG to text.

What your transcript looks like

Every transcript includes segment-level timestamps and speaker labels by default. Word-level timestamps are available on paid plans. Below is a real (anonymized) excerpt of what you'll see in the editor and in the exported TXT/DOCX.

[00:00:03] Speaker 1: So the thing about video transcription is
[00:00:06] Speaker 1: most people don't realize it's just audio underneath.
[00:00:11] Speaker 2: Right, the container format doesn't really matter.
[00:00:14] Speaker 2: MP4, MOV, MKV — they're all wrappers.
[00:00:19] Speaker 1: Exactly. Whisper only sees the audio track.

Prefer plain text without timestamps or speaker labels? Toggle them off before export — the DOCX and TXT outputs can be paragraph-only if that's what you need for a blog post or article draft. SRT and VTT preserve timing by definition (they're subtitle formats). JSON keeps the raw word-level timing array for developers building custom video editors or search interfaces.

Language auto-detection

The language dropdown pre-selects based on your browser locale. A Turkish speaker on a Turkish browser sees Turkish selected by default; a Portuguese speaker sees Portuguese; a Chinese speaker sees Mandarin. This is deliberate, not cosmetic — it fixes a bug that affects most third-party transcript tools built on Supadata's caption API.

The bug: when a video has multiple auto-generated caption tracks (TikTok and Instagram often have three or four — Vietnamese, Indonesian, English, plus the original language), a tool that asks the caption API for "auto" can get a random language back. That's where the notorious "why is my English video's transcript in Spanish" complaint comes from. Pre-selecting the user's actual browser language cuts the miss rate significantly.

All 99 Whisper Large-v3 languages are supported. Non-Latin scripts (Chinese, Japanese, Arabic, Hindi, Thai) render cleanly in the editor and every export format. Right-to-left scripts (Arabic, Hebrew, Persian) preserve reading direction in DOCX and TXT. Auto-detection doesn't work perfectly on very short clips (under 3 seconds) — for a 5-second Reel, override manually if needed.

Accuracy by content type

Accuracy varies with audio quality, not video quality. A 4K studio production with laptop-mic audio is worse than a 480p phone recording with a lavalier close to the speaker. Numbers below are typical ranges from our internal testing across ~800 videos and match the general findings on the Hugging Face Open ASR Leaderboard for Whisper Large-v3.

Proper nouns — brand names, product SKUs, technical jargon, foreign names — miss more often than regular vocabulary (20–30% error rate even on otherwise clean audio). The editor's search-and-replace fixes all instances in one pass; that's where most of the editing time goes.

Limits, pricing, and the free preview

Honest limits, stated upfront so nothing surprises you at signup:

• ●Free preview: first 200 characters of any transcript, no signup, no card. Enough to confirm language, quality, and speaker structure.
• ●Free account: 30 minutes/month of full transcription with all features. No credit card required. Enough for ten 3-minute TikToks, two 15-minute podcasts, or one 30-minute lecture per month.
• ●Paid plans: start at $2/month for 200 minutes; details at /pricing.
• ●Max file size: 500 MB per upload.
• ●Max duration: 4 hours per video.
• ●URL fetches: we time out at 30 seconds. If a video takes longer to fetch (large Vimeo files, slow CDN), download it and upload the file instead.

Roughly $0.01 per transcribed minute on the Pro plan — cheaper than typical per-minute transcription services at $0.10–$0.25/min, and 100–500× cheaper than human transcription at $1.50–$2.50/min. Human transcription is still the standard for legal filings, broadcast captions requiring ADA-compliant certification, and other cases where a stamp of certification matters — for research, notes, SEO content, subtitles, and social clips, AI is fine.

This tool vs the YouTube-specific tool

We run two tools that share the same Whisper backend but differ in the front door:

Both tools produce the same downstream output (same editor, same export formats, same account). Paste a YouTube URL here by mistake and we'll suggest the YouTube tool — no data loss.

Common workflows

What people actually do with the transcript once it's generated:

When to use a different tool

This page is the generic uploader — good for most cases. A few specialized tools handle their niche better:

Methodology and disclosure

Accuracy numbers: derived from internal testing across ~800 videos spanning single-speaker explainers, podcast recordings, Zoom meetings, TikTok/Instagram Reels, lectures, and vlogs. Whisper Large-v3 baseline accuracy is consistent with published results on the Hugging Face Open ASR Leaderboard (verified July 2026).

Whisper model documentation: OpenAI Whisper announcement and the Whisper GitHub repo (verified July 2026).

Human transcription rate reference: Rev.com published pricing at $1.25/min for AI + human review, ~$1.50/min for standard human transcription (verified July 2026). Numbers may drift; the order of magnitude is the point.

Editorial standards: we don't claim "100% accuracy" or "free forever." The preview is free, the free account has a monthly cap, and paid plans start at $2/mo. Full disclosure at /about/editorial-standards.