sqwartHome

Audio to Text Converter

Turn any recording into text with Whisper AI, transcribed on your own device.

  • Free
  • No upload
  • No watermark
  • No sign-up
  • AI runs on your device
How it works

How to use the audio to text tool

  1. 1

    Choose an audio file (MP3, WAV, M4A, OGG, FLAC) or a video.

  2. 2

    Pick the spoken language, or leave it on automatic, and a model: Fast, Balanced or Accurate.

  3. 3

    Press Transcribe and watch the text appear as Whisper listens.

  4. 4

    Copy the text or download it as TXT, TXT with timestamps, or SRT.

In detail

How it works on your device

The tool runs Whisper, the open speech recognition model OpenAI trained on 680,000 hours of audio, inside your browser. Your file is decoded to 16 kHz mono audio on your computer, cut at quiet moments into pieces shorter than 30 seconds (Whisper's listening window), and each piece is written out with start and end times. Text appears as each piece finishes, so you can read along while it works.

Nothing is uploaded: the audio never leaves the tab. The first time, your browser downloads the model from Hugging Face (64 MB for Fast, 136 MB for Balanced, 250 MB for Accurate on a processor; a little more on a graphics card) and keeps it, so the next file starts at once. Where your browser has WebGPU (recent Chrome and Edge on most laptops and desktops) the model runs on the graphics card; everywhere else it runs on the processor with WebAssembly, which is slower but works.

Accuracy, speed and limits

  • Fast (Whisper tiny) suits clear English speech; Balanced (base) is a good default; Accurate (small) is the one to use for accents, noisy rooms and for Hindi, Tamil, Bengali and other Indian languages, where the smaller models make many mistakes.
  • On a typical laptop processor, Balanced transcribes a 10-minute file in roughly 3 to 5 minutes; Accurate takes about as long as the audio. With WebGPU it is usually several times faster.
  • Whisper understands about 99 languages. You can pick the language or let it guess from the first words; picking it is more reliable, especially for mixed Hindi and English.
  • It does not tell speakers apart, and it can mishear names, numbers and jargon. Read the text through before you publish it.
  • Very long files need memory: about 60 MB per hour of audio for the samples, plus the model. Phones may run out of memory on long recordings; a laptop is the safer choice.

What you get

A clean transcript in paragraphs, split wherever the speaker pauses for two seconds or more, ready to paste into a document, a blog post or show notes. You can also save it with a timestamp on every line, which is handy for interviews and lectures, or as an SRT subtitle file. For subtitles with line lengths set for video, use the auto subtitle generator.

FAQ

Questions, answered.

Is it really free, with no limit on minutes?

Yes. The model runs on your own device, so it costs us nothing per minute and there is no sign-up, watermark or daily cap. The only limit is your device's memory and patience on very long files.

Is my audio uploaded anywhere?

No. The file is read and transcribed inside your browser tab. The only download is the Whisper model itself, from Hugging Face, the first time you use it.

Which languages does it support?

Whisper was trained on about 99 languages, including English, Hindi, Urdu, Bengali, Tamil, Telugu, Marathi, Spanish, French, German, Arabic, Japanese and Chinese. Accuracy is best in English and the major European languages; for Indian languages choose the Accurate model.

Why is it slow on my computer?

Without WebGPU, the model runs on the processor. Chrome or Edge on a laptop with a recent graphics chip is several times faster. The Fast model is quickest, at some cost in accuracy.

Can it translate as it transcribes?

Yes, into English only. Turn on Translate to English and Whisper writes English text for speech in any language it knows.

Which audio formats work?

Anything your browser can play (MP3, WAV, M4A, AAC, OGG, Opus, FLAC, WebM) is read directly. Other formats, and very large files, are converted with ffmpeg in your browser first, which adds a one-time 31 MB download.

Keep going

More free audio tools