Stays on your device. WebGPU (or WebAssembly fallback) runs Whisper in your browser. No upload, no signup, unlimited use.
Convert MP3 to text without handing over the file
Almost every "MP3 to text" service works the same way: you upload the file, it lands on somebody's server, a model runs there, and you get text back, usually after a free tier runs out. The upload is the part people stop and think about, because an MP3 is rarely just an MP3. It is a client call, a therapy session, a lecture you were allowed to record, a song that is not out yet.
The tool above removes that step entirely. It downloads OpenAI's Whisper model into your browser once, then reads your MP3 straight off your disk and transcribes it on your own processor. No upload, no account, no minute counter. The audio has no reason to leave the machine it is already on, so it does not.
How to convert an MP3
- Drop the MP3 onto the panel above, or use the file picker. Nothing uploads when you do this; the browser is just reading the file.
- The first run downloads the Whisper model, which takes a moment on a normal connection. Your browser caches it, so every MP3 after the first starts immediately, and it keeps working with no connection at all.
- Watch the transcript build as it goes. Longer files stream in section by section rather than making you wait for the whole thing.
- Download plain text for reading, SRT or VTT if the audio belongs to a video, or JSON if you are going to process the timestamps yourself.
Where an on-device transcript is the only sensible option
Some recordings simply should not be uploaded, and no privacy policy fixes that. Interview audio under embargo, anything covered by a nondisclosure agreement, medical or legal conversations, HR recordings, unreleased music, and research data governed by an ethics approval all fall into the category where "we delete it after 30 days" is not an answer you can act on.
The distinction worth internalising is that privacy here is a mechanism rather than a promise. A cloud transcriber asks you to trust its data handling. An on-device model gives you nothing to trust, because there is no server in the loop to trust in the first place.
Frequently asked questions
- How long does an MP3 take to transcribe?
- It depends on your hardware, not on the file. On a machine with WebGPU a one hour MP3 typically finishes in a few minutes, because the model processes the audio far faster than real time. On an older laptop falling back to WebAssembly it is slower, and the progress bar will tell you where it is. There is no queue and no server load, so the number is yours alone.
- Is there a file size or length limit?
- Files up to 500 MB, with no cap on duration. That is a browser memory limit rather than a policy, and 500 MB of MP3 is roughly eight hours of speech at a typical bitrate. Nobody is metering minutes, because there is no server doing the work to bill you for.
- Does it work with formats other than MP3?
- Yes. WAV, M4A, FLAC, OGG, OPUS and WebM all work, as do video containers like MP4 and MOV, because the browser decodes the audio track out of them first. If your recorder saved a voice memo as M4A, you do not need to convert it to MP3 first.
- Is my MP3 uploaded anywhere?
- No. The file is read by your browser and decoded in memory on your own machine. There is no upload step and no copy on a server, which you can confirm the direct way: transcribe once so the model caches, turn off your wifi, and transcribe again. It still works, because there was never a network round trip doing the transcription.
- How accurate is it compared to a paid service?
- It runs OpenAI's Whisper, which is the same open source model family that a lot of paid transcription products wrap and resell. On clear speech it is very good. It degrades the way every speech model does: heavy overlap between speakers, thick background noise, and very quiet recordings all cost accuracy. It does not label who is speaking.
- What do I get out at the end?
- Plain text, an SRT or VTT subtitle file with real timecodes, or JSON with per-segment timestamps if you want to post-process it. All four are free and download straight from the page.