DeTranslator: How It Works (Explainer & FAQ)

What DeTranslator does

DeTranslator removes the translated passages from an audio or video recording, leaving the original speaker clean and intact.

To be precise about what the output looks like: the translated passages are cut out entirely. The interpreter's audio and, for video, the matching picture are removed together, so you get back one continuous recording that is shorter than the input by the length of the removed passages. Nothing is muted in place: there are no silent gaps and no frozen or blacked-out video where the interpreter used to be.

The problem is a specific and common one: a recording (a sermon, a lecture, a conference talk, a dharma talk) has a live interpreter, or several interpreters into different languages, who take turns with the original speaker, translating each passage in the speaker's pauses, so every voice ends up on the same audio track. Removing the interpreted passages by hand means manually finding and cutting each one, which can take hours for a single recording. People who search for a way to do this automatically are consistently told it's "near-impossible once mixed into one track," and no off-the-shelf tool does it. DeTranslator automates that removal.

Will this work on my recording?

DeTranslator works when the voices take turns. That is the usual pattern in consecutive interpretation: the speaker talks, pauses, and an interpreter renders that passage in another language before the speaker continues. We find each interpreted passage, whichever language and whichever interpreter it belongs to, and cut it out on the timeline, leaving the original speaker continuous.

It does not work on a dubbed or voice-over recording, where another voice is mixed on top of the original speaker and the two are talking at the same time. DeTranslator cuts on the timeline; it does not separate voices that occupy the same moment. On that kind of recording, cutting a passage would either leave the voice-over in place or delete the original speech underneath it, so the result would not be what you want. If your recording is a dub or voice-over like that, this is not the right tool and you should not upload it.

A little overlap where the voices briefly cross is fine and expected: the design keeps the original speaker intact even if that means a short residual sliver of an interpreter's voice survives at an overlap. The line is simple: turn-taking works; a full simultaneous dub or voice-over does not.

Getting started

DeTranslator is live: create a free account, upload your file, and you will see the exact price for that file before you are asked to pay anything. See How it works for the full step-by-step flow.

Pricing: how a quote is generated

The price is based on the duration of your uploaded file, calculated as:

quote = the greater of $5.00, or $29.00 per hour of input audio/video duration (rounded up to the nearest cent)

In plain terms: we charge $29.00 per hour of the file's length, with a $5.00 minimum charge on any job. A few examples:

This formula is fixed and applies to every customer: there is no negotiation and no hidden fee. You see the quote before you commit to anything, based on your file's actual duration.

How capture works: when you confirm a job, a hold is placed on your card for the quoted amount and you are not charged yet. The charge is only captured after your file is processed successfully, and the amount captured is never more than the quote. If processing fails, the hold is released and you are not charged.

Turnaround expectations

In our own testing, the pipeline processes a 2-hour recording in roughly 1 hour. That is the only figure we have measured so far, and we only publish numbers we have actually measured.

We do not yet promise a guaranteed turnaround time for the live service: typical wait times depend on queueing and load, and not enough real jobs have run for us to state one honestly. As they do, we will publish measured figures here. If your job is time-sensitive, email support@detranslator.com before uploading and we will tell you what to expect.

The quality promise

DeTranslator is built around retention-first removal: the design goal is to keep speech in the language you choose fully intact, and to find and remove all speech that is not in that language. Any number of interpreters can be involved: a talk relayed turn by turn into French and German, for example, comes out as the speaker's own language alone in one pass. Retention comes first, even at the cost of an occasional small residual sliver of non-target speech surviving in genuinely overlapping speech.

The pipeline is designed to compute an automated per-job "leak score": an extra check that re-transcribes the output and measures how much non-target speech was left behind, and to store that score alongside the job. The intent is for this score to back a quality gate (for example, automatic retry, or flagging a job for a refund if the score comes back bad).

Important: this is a policy intent, not a shipped guarantee today. An automated quality measurement is recorded for every job, but it does not yet automatically gate capture, trigger a retry, or trigger a refund. Until it does, we are not advertising a "verified clean or refunded" guarantee as an active, automatic policy. If a result is materially defective, our Terms of Service already cover re-processing or a refund on a case-by-case basis: contact us within 14 days of your job completing.

Supported formats

Uploads are audio or video files, up to 6 hours long per file. Most common audio and video formats work. We have not yet published a definitive list of file extensions and codecs; if yours is unusual, try it (you only ever pay after your file has been processed successfully) or email support@detranslator.com first and we will check.

Supported languages to keep

You pick the language you want to keep from a list of 73 languages built into the product today: Afrikaans, Albanian, Arabic, Armenian, Azerbaijani, Basque, Belarusian, Bengali, Bosnian, Bulgarian, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, Ganda, Georgian, German, Greek, Gujarati, Hebrew, Hindi, Hungarian, Icelandic, Indonesian, Irish, Italian, Japanese, Kazakh, Korean, Latvian, Lithuanian, Macedonian, Malay, Maori, Marathi, Mongolian, Norwegian (Bokmål and Nynorsk), Persian, Polish, Portuguese, Punjabi, Romanian, Russian, Serbian, Shona, Slovak, Slovenian, Somali, Sotho, Spanish, Swahili, Swedish, Tagalog, Tamil, Telugu, Thai, Tsonga, Tswana, Turkish, Ukrainian, Urdu, Vietnamese, Welsh, Xhosa, Yoruba, and Zulu. English, German, French, Spanish, Italian, Portuguese, Dutch, and Russian are shown first in the upload form as the most common choices; the rest are listed alphabetically.

Data handling and privacy

Processing a file requires sending audio to our transcription partner AssemblyAI as part of the pipeline. Uploaded files and processed results are automatically deleted 30 days after your job completes; job metadata (without the media) is kept for accounting. See the full Privacy Policy and Terms of Service for the complete detail, including how to request deletion of your account and data.

How to get started

Create a free account, upload your file, and you will see the exact price for that file, calculated from its duration, before you are asked to pay anything. You are only charged after your file has been processed successfully.