SubtitleLens

Measured, not claimed

0.37%

The error rate we measured, not the one we were told.

One wrong word in 267, from a transcription engine we ran against a text whose wording was known before the tool ever saw it. The same score came back after we degraded the audio to laptop-mic quality.

Happy Scribe · 2-minute clip · scored against Project Gutenberg · read the method

What a measurement looks like here

Not an adjective about how natural a dub sounds. The actual speech blocks, read off the two audio files.

English original

Chinese dub

0s9s18s

Speech blocks in an 18-second clip and in its Chinese dub. Two of the five onsets land within 2 ms and 14 ms of the original speaker. The Chinese runs 14.3 % longer, so the dub absorbs the difference by eating 33 % of the pauses. How this was measured

Four more numbers

Free tools, run in your browser

Your file is parsed locally and never uploaded to us. No account, no quota, no watermark.

SRT translator

Translate .srt and .vtt into 30+ languages. Bilingual output for study.

SRT ⇄ VTT converter

Swap between the two formats. Timestamps convert 1:1.

Subtitle shifter

Move every cue earlier or later to fix a constant offset.

Latest from the archive

Every review and guide →