Skip to the converter
InBrowserConvert

0 bytes uploaded

Files stay on this device

Audio to Text Converter (Private, No Upload)

An audio to text converter for an MP3, WAV, or M4A. Transcription runs on this device after a one-time download of the speech model’s weights. There is no account. A long recording is limited by this device’s memory, not by a quota.

Whisper tiny is about 41 MB of weights (10 MB encoder and 31 MB decoder). Whisper base is about 77 MB (23 MB encoder and 54 MB decoder). The browser downloads the one you pick from Hugging Face the first time, then keeps it in its cache. That download is the model weights, not your file. A speech engine may also load from a CDN.

Model
Save as

The model download starts when you transcribe. WebGPU is used when this browser has it. Otherwise the model runs in WebAssembly. iPhone needs Safari 26. Cancel closes the transcriber.

InBrowserConvert Pro (coming soon): batch, bigger files, no limits, still 100% on-device. $29 lifetime or $19/yr.

A transcript that stays on the device

Audio to text here means the browser runs Whisper. Some transcription sites take the recording onto their server or stop at a free-tier length. Here the file stays on the device, the text has no watermark, and there is no account. Memory is the limit.

Names, overlapping voices, and background noise still come out wrong sometimes. Read the text before you rely on it. The page does not claim a perfect transcript.

What the first download is

Whisper tiny is about 41 MB of weights. Base is about 77 MB. Progress during that step names the model, so a network entry for huggingface.co is the weights. The upload meter counts bytes this page sends, and the recording is not one of them.

How to convert

  1. 1. Add an MP3, WAV, or M4A

    Drop the recording or choose it on this phone or computer. The picker hands the file to the page. It is not sent to a server.

  2. 2. Pick the model and the language

    Tiny is about 41 MB of weights. Base is about 77 MB and is slower. Leave the language on Auto, or choose one. Plain text is the default. SRT and VTT are there when you want timestamps. The weight download starts when you press Transcribe audio.

  3. 3. Download the text

    The file has no watermark. A phone keeps the decoded audio around 160 MB, and a computer around 400 MB. Cancel closes the transcriber. One file runs at a time. iPhone needs Safari 26 to read the recording. WebGPU is used when the browser has it, and WebAssembly covers the rest.

Questions

Is this an audio to text converter, free?

Yes. The page is free. The converter runs in the browser on an MP3, WAV, or M4A. The recording stays on this device. There is no watermark.

Can I transcribe audio to text free?

Yes. Press Transcribe audio and the text is written in this tab. The first run downloads the model weights from Hugging Face. That transfer is the model, not your recording.

Is audio to text free with no sign-up?

Yes. There is no account and no login. The page does not ask for an email to download the text. The Pro note on this page only remembers a plan click in this browser.

Is audio to text free and unlimited?

There is no account cap and no daily quota. The recording still has to fit in this device’s memory. A phone stops around 160 MB and a computer around 400 MB. A longer file is refused before transcription starts, and it is not uploaded.

Can I transcribe an audio file to text?

Yes, for MP3, WAV, and M4A when this browser can decode them. Chrome and Edge can. iPhone needs Safari 26. If a file cannot be read, the page says so and keeps it on the device. You can save plain text, SRT, or VTT.

Does it work on iPhone?

In Safari 26 or newer, which can decode the recording. The model then runs in the browser, with WebGPU when Safari exposes it and WebAssembly otherwise. Older iOS shows an error and does not upload the file. A phone also uses the smaller ceiling, about 160 MB.

Which files does this page actually accept?

MP3, WAV, and M4A. The download is TXT, SRT, or VTT. Other audio types are refused, and they are not uploaded. A video file belongs on the video to SRT page.

Where do the model weights come from?

From Hugging Face, the first time you transcribe: onnx-community/whisper-tiny or onnx-community/whisper-base, quantized. The speech engine’s WASM file can load from jsDelivr. This site does not host those files, and neither download is your recording. The browser cache keeps the weights for the next visit.

Related tools