Transcribe audio to text on your computer

A transcript without sending your recording anywhere

Interviews, lectures, meetings, podcasts, voice notes: recordings are often private. Wolfcaps transcribes them on your own computer. The speech recognition model runs in your browser on your GPU, so the audio is never uploaded. It's free, and you don't need an account.

You get a readable transcript in paragraphs. Every sentence stays linked to its moment in the recording, so you can click it to hear it again.

How to transcribe audio with Wolfcaps

1. 1Open Audio → transcript and drop your file: MP3, M4A, AAC, WAV, FLAC, OGG, Opus or WMA. A video file works too. If the recording is on your phone, click Send from your phone and scan the QR code.
2. 2In Transcribe this recording?, choose the spoken language (English, Hebrew or Russian) and Balanced or Best. The model downloads once, the first time, and stays in your browser.
3. 3Wolfcaps shows how long it will take on your device, then starts. The transcript appears as it's written.
4. 4Press play to follow along: the current sentence is highlighted. Click a sentence to hear it and fix it in place. Enter saves and moves to the next sentence; Esc cancels.
5. 5Click Export… and save a TXT file, or a subtitle file with timings.

What you get

• TXT.

  The words only, in paragraphs, for notes, documents or pasting elsewhere.

• SRT, VTT or ASS.

  The same text with timings, if you need subtitles. The SRT generator page explains the formats.

• A translation.

  Click Translate… to translate the transcript into one of 20 languages. See translate subtitles for how it works.

If you'd rather fix the text line by line, open the Full editor. It has find & replace, timing on a waveform, split and merge, and undo.

Hebrew transcription

Hebrew speech is transcribed with ivrit.ai's Hebrew Whisper models, which we converted to run in the browser. You choose between two levels:

• Balanced.

  The faster one: about a 1.6 GB download.

• Best.

  A bigger model: much more accurate on hard audio, and about 3 times slower. About a 2.1 GB download.

Start with Balanced. If the recording is noisy, fast or hard to follow, try Best.

English and Russian work the same way, with their own Balanced and Best choices.

Long recordings

A two-hour lecture or a long interview is fine: multi-hour files work. On a recent desktop GPU, Wolfcaps runs about 10 times faster than real time, so a 10-minute English recording takes about a minute. Slower GPUs take longer. Before it starts, Wolfcaps runs a short speed test and shows an estimate for your device.

If you stop partway, keep the project when Wolfcaps asks. It waits under Recent projects in your browser and continues from where it stopped.

What it can't do yet

• Speech recognition works for English, Hebrew and Russian only. Recordings in other languages can't be transcribed yet.
• The transcript doesn't say who is speaking. You get the words, in paragraphs, without speaker names.
• You need a computer with Chrome, Edge, Brave or another Chromium browser, or Safari 26 or newer on a Mac, and a GPU the browser can use (WebGPU). Firefox, iPhone and iPad aren't supported yet.