Prepare your transcripts
Two transcripts. One short review queue.
You need a transcript of the original speech and a separate transcript of what the dubbed voice actually says. A translated script alone cannot show whether the recorded dub changed a number. This tool checks written digits and supported units in English, Spanish and German number formats.
For more context, explore the dubbing review guides: collecting transcripts, worked numeric examples and a pre-publish checklist.
1. Get the words from both tracks
Use the audio or video files you are authorized to review. Export timestamped captions from your editor or transcription tool separately for the original and the actual dubbed audio. Check that you selected the intended language track. If you only have a translated script, transcribe the recorded dub first using a tool you trust with that material. This website does not download videos or transcribe audio.
Listen to a few passages, especially numbers, to check that each transcript matches its own recording. Keep original timestamps and meaningful subtitle boundaries. Do not replace written-out number words with digits unless you have verified the speech: number words are outside this check.
2. Export a supported format
Use UTF-8 SRT, WebVTT (.vtt), or timestamped JSON, up to 2 MB per file. Each transcript may contain at most 4,000 segments and 2,000 numeric occurrences. Every segment needs non-empty text, a nonnegative start time and an end after its start. Plain text without timestamps is not supported.
1 00:00:12,000 --> 00:00:17,000 Start with 120 grams of protein.
JSON uses seconds, not milliseconds. Arrays with dur instead of end and Whisper-style {"segments": [...]} objects also work. Optional confidence must be a number from 0 to 1; leave it out if your transcription tool does not supply a suitable value.
[{"start":12,"end":17,"text":"Start with 120 grams of protein."}] 3. Try the practice files
These fictional transcripts contain two deliberate substitutions. They have no associated audio and are a functional demonstration, not evidence of real-world accuracy.
Download original SRTDownload dubbed SRT
Open the checker, import the original file on the left and dubbed file on the right. Select English for the original and Spanish for the dub; leave the offset at 0. The built-in Load sample button loads the same pair.
- Expected coverage: 4 numeric occurrences per track, 2 matched pairs, 2 review items.
- At original 01:12 / dub 01:13, 120 grams becomes 240 grams.
- At original 01:38 / dub 01:39, 29 dollars becomes 290 dollars.
- 150 pounds and 68 kilograms match as a supported rounded conversion.
4. Align and review your own files
Choose the number format used in each transcript: English reads 1,500.5; Spanish and German read 1.500,5. If the dub consistently starts 30 seconds later, enter +30 as its offset. If it starts earlier, use a negative offset. The offset range is −300 to +300 seconds. A single offset cannot correct timing drift that changes throughout the recording; split or realign those clips first.
Findings show track-specific timestamps. Play the original and dub at their respective times in your own editor, verify the words and decide whether to correct the dub, the transcript or the timing. Nearby unmatched values can share a review item; they are not a confirmed semantic replacement. Export the JSON report before closing the page.
If something looks wrong
- Timestamp error: export captions with timing, check the example above, and ensure each end follows its start.
- No supported numeric occurrences: the transcript may contain only number words. An empty report does not mean the dub is correct.
- Many timing warnings: check track selection, clip coverage and offset.
- Unexpected flags: inspect decimal separators, transcription errors, intended step renumbering and unsupported units.
- Matching numbers but wrong meaning: semantic role swaps, translation quality and audio quality still require human review.
Imported transcripts and reports stay in your browser. Anonymous usage counting can be disabled in Privacy. For help, use Contact and describe the problem without including private transcript content.