How it works
From a recording with the speaker and interpretation mixed together to a clean recording of the original speaker. Here is what the service does, step by step.
The problem it solves
When a talk is interpreted live and the interpreter takes turns with the speaker (consecutive interpretation), the interpreter's translated passages land in the same audio track as the original. Standard audio tools cannot pull one speaker out of a single mixed track, and text-based editors still make you find and delete every translated passage by hand. DeTranslator automates finding and cutting the interpreter's passages.
This works when the voices take turns, however many interpreters are involved. It does not recover the original speech from under a dub or voice-over, where another voice talks over the speaker at the same time, because the cuts are made on the timeline and cannot separate voices in the same moment. The FAQ explains where it fits and where it does not.
Step by step
1. Create a free account
Sign up with an email address, or continue with your Google account.
2. Upload your file
Upload the audio or video recording that has the interpreter's voice mixed in, and pick the language you want to keep from a list of 73 supported languages.
3. See your price
The price is calculated from the file's duration: $29.00 per hour, with a $5.00 minimum. You see the quote before you commit to anything.
4. We process it
The pipeline detects the interpreter's passages and cuts them out of the recording entirely, audio and matching video together. What remains is a shorter, continuous recording of just the original speaker, untouched.
5. You are charged only on success
A hold is placed at your quoted price when you confirm the job. It is captured only after processing succeeds, never more than the quote. If processing fails, the hold is released and you pay nothing.
6. You download the result
You get a link by email to download the cleaned recording: just the original speaker, with the interpreter's passages cut out. Your download is shorter than your upload by exactly the material that was removed.
The quality approach
The design is retention-first: keep the original speaker fully intact, never cutting the voice you want to keep, even at the cost of a small residual sliver of non-target speech at a genuine overlap. A per-job automated quality check (a "leak score") is on the roadmap but is not a shipped guarantee today. The FAQ spells out exactly what is built versus planned.