Translate Italian video to English
Shycut’s Accurate model listens to Italian and writes English, at 17.8 BLEU in OpenAI’s test. The Fast model manages 11.3, so translation runs on Accurate.
Drop your video here
Choose your videoMP4, MOV, WebM or MKV. It opens right here, on your device.
- Nothing uploaded
- No sign-up
- No watermark
How well it translates Italian into English
| Model in Shycut | BLEU into English | From the paper |
|---|---|---|
| Accurate (Whisper small) | 17.8 | Table 14, “Whisper small” row, Italian column · OpenAI |
| Fast (Whisper base) | 11.3 | Table 14, “Whisper base” row, Italian column · OpenAI |
BLEU compares a machine translation with human ones; higher is closer. These are OpenAI’s scores on the FLEURS set.
Accurate scores 1.6 times as high as Fast here, so a translate link opens Shycut with Accurate chosen.
How often it gets a Italian word wrong
| Model in Shycut | Word error rate | Roughly | From the paper |
|---|---|---|---|
| Accurate (Whisper small) | 9.8% | 1 word in 10 | Table 13, “Whisper small” row, Italian column · OpenAI |
| Fast (Whisper base) | 17.9% | 1 word in 6 | Table 13, “Whisper base” row, Italian column · OpenAI |
Word error rate is the share of words a model gets wrong (swapped, missed or added) against a human transcript, in OpenAI’s own test on the FLEURS speech set; lower is better.
Shycut runs these two models converted to work in a browser (with 8-bit or 4-bit weights on some devices), so its results can differ a little from the paper’s.
How the translation is made
- While it listens
- The speech model writes English as it hears the Italian: no separate translation step, and nothing is sent anywhere.
- Timed to the speech
- Each line follows the speech it translates; the timing of single words is approximate, since English word order differs.
- Kept apart from the script
- English words aren’t the ones said, so Shycut doesn’t use them to cut the video or find hesitations; the Text tab offers to write the speech as spoken.
Italian into English, checked by you
Italian ranks sixth of the 9 languages Shycut translates. A model translation is a good first draft: keep the lines that read well and rewrite the ones that don’t, line by line in the Captions tab.
The Italian itself is heard with about 1 word in 10 wrong. Where a word was misheard, the English will be off too, so a quick listen while reading pays off.
How to do it in Shycut
Open your video
The button opens Shycut set to translate Italian speech into English, with the Accurate model chosen.
Write the English captions
The model listens to the speech and writes English lines, timed to it, on your device.
Check, then save
Read the lines through, correct what needs it, then save the video with them or download an SRT file.
Where these facts come from
Questions
How accurate is Italian to English?
17.8 BLEU for Accurate and 11.3 for Fast, in OpenAI’s figures.
Can I switch back to Italian captions?
Yes. Choose “As spoken” in the Captions tab and write them again.
Is the video uploaded?
No. Everything runs on your device.
Your video stays with you.
Open Shycut, drop the video in and export. Free, with no account, and your video never leaves your device.
Choose your video