Transcribe audio to text

Audio to text in a few clicks

Choose a recording and press Transcribe: an AI speech-recognition model listens to it on your device and writes down what is said. Copy the text straight from the page or download it as a text file, or as subtitles with timings if you need them.

It handles MP3, WAV, M4A, FLAC, Ogg and Opus files, voice notes from phones and messaging apps, and the sound of video files. The language is detected automatically; if a recording starts with music or silence, choose the language yourself for a better start.

Good to know

Accuracy is highest for clear speech, one person at a time, close to the microphone. Interviews and meetings work well; noisy rooms, music and people talking over each other make mistakes more likely, so give the text a read before you publish it.

Private by design

Most transcription sites upload your audio to a server and charge per minute. Here the model runs in your browser, so interviews, medical notes and private messages stay on your device. The first time, the model (about 136 MB) is downloaded and kept by your browser. Recordings up to 90 minutes.

The sound is converted to 16,000 samples per second and split into 30-second windows. Each window becomes a spectrogram (how loud each pitch is over time) that the Whisper base model, an open-source speech-recognition network by OpenAI (MIT licence), reads to write the words with timestamps. Windows overlap where a sentence is cut, so no words are lost between them.