Hear your recording in another language

Attach audio or video — or paste a link to one — and say which language you want. It transcribes, translates for the ear rather than the page, and speaks it back.

Translate a recordingFree to try. Nothing to install.

How it works

Three steps, and none of them is “configure”.

01
Give it the recording
Drop a file anywhere on the page, or give it a URL and it fetches the file itself rather than asking you to download and re-upload.
02
You see the transcript first
Everything after it inherits its mistakes, and you are the only one who can hear whether it is right. So it shows you the transcript before it translates anything.
03
It speaks the translation back
In a voice matched to the target language, with the transcript and the translation kept as files you can correct and ask it to say again.

What it can do

Three kinds of model in one run
Transcription, then a chat model translating, then speech. You pick the chat model; the voice is chosen to match the language it is reading.
Translated for the ear
Contractions, natural word order, the register the speaker actually used. A translation that reads correctly and sounds like a form is the wrong output for this job.
It flags what it had to decide
Names, jokes and idioms are where a translation makes choices. It says which ones it was unsure about instead of handing you a smooth result that hides them.

How it is set up

The mechanics, so you know what you are getting before you sign in.

Environment: A Linux container with curl and your workspace mounted, so a recording it downloads lands where the transcription tool can read it.
Tools: Transcribe, generate speech, read and write files, glob, and a shell.
Model: Any chat model for the translation. It looks the speech model up against the language you asked for, and names the voice it is using before it starts.
Starting files: Nothing seeded: the input is a recording of your own, and a sample clip would demonstrate the pipeline while answering a question nobody asked.
What persists: The audio, the transcript and the translation stay in your workspace after the conversation ends.

Things people ask it

“Transcribe my recording and give it back to me in Spanish.”
“Which languages can you actually speak, and what happens if I ask for another?”
“Here is a link to the interview — translate it into German and read it back to me.”
“Just the transcript this time, with the speakers separated.”

What it will not do

Every one of these is a real constraint we have hit, not a roadmap item.

Four languages it can speak, not four hundred. Ask for another and it says so up front and offers the written translation, rather than spending the run and telling you at the end.
25 MB is the transcription limit. It checks the file size before spending a call on it.
There is no ffmpeg in the sandbox, so it cannot convert or re-encode audio. It usually does not need to — mp4 and webm go straight in.

Which model does this best

Measured over real runs of this agent, per model. A run counts as finished when it produced the answer on its own — nothing failed, nobody was asked to approve anything, and it did not run out of steps. Read all three columns together: a model that finishes fast by giving up scores badly on the first, and one that completes everything by grinding scores badly on the second.

Model	Completion rate	Median cost	Median time	Runs
claude-sonnet-5anthropic	100%	$0.014	9s	4

Completion rate: Runs that finished the job without you having to step in.
Median cost: What a typical run costs, in credits.
Median time: Wall clock, from the first message to the answer.

Medians over the runs behind each row. A model appears once it has 3 runs on this agent, and the run count is shown so you can judge how much a figure rests on.