Transcribe audio and video to text free

How to use it

1. 1

   Add your recording. Drop an audio or video file. Phone voice memos (M4A), MP3, WAV and MP4 all work.

2. 2

   Choose the language and speed. Leave the language on automatic or pick it for better accuracy. Balanced suits most recordings; choose Most accurate for noisy audio or accents.

3. 3

   Copy or download the text. Tick Add timestamps if you want to find quotes later. Fix any names the AI misheard, then copy or save it.

How accurate is it?

Whisper is one of the most accurate open speech models available. On clear audio with one speaker it usually gets well over nine words in ten right. Accuracy drops with background noise, crosstalk, strong accents and specialist vocabulary. Names, brands and technical terms are the most common mistakes, so always skim the transcript before you use it.

Tips for better transcripts

• Pick the spoken language instead of auto-detect when you know it.
• Use the Most accurate model for noisy rooms, phone calls and non-native speakers.
• Record close to the mic. Distance hurts accuracy more than a cheap mic does.
• Split very long recordings (2 hours+) into parts if your device is low on memory.

Private by design

Transcription services usually upload your audio to their servers, which is a problem for interviews, medical notes or anything confidential. Here, the Whisper model downloads to your browser once and runs on your own device. Your audio file is decoded and transcribed locally and never leaves your computer or phone.

How long does it take?

On a recent laptop with a modern browser, the Balanced model typically transcribes faster than real time. Phones and older computers are slower. Speed depends on whether your browser supports WebGPU (Chrome and Edge do), which lets the model use your graphics chip.

Questions