Fala para texto (no dispositivo)

Transcreva fala de um arquivo ou do microfone com um modelo Whisper no dispositivo. O áudio nunca sai do seu dispositivo; o modelo é baixado no primeiro uso.

First use downloads the selected Whisper model from Hugging Face (see sizes in the model list). After that, transcription runs entirely on your device: your audio is never uploaded anywhere — only the model files are downloaded.

Model

Language

Audio file

…or record from the microphone

Include timestamps

Como usar

Choose a model (Tiny is fast, Base is better) and the audio language.
Pick an audio file, or record from the microphone.
Press Transcribe. The first run downloads the model.

Observações

Transcription runs a Whisper model on-device with Transformers.js. Your audio is never sent to a server; only the model files are downloaded. Audio is converted to 16 kHz mono in the browser before it reaches the model. Accuracy varies a lot with recording quality, speakers and jargon — treat the output as a draft that needs human review for anything important.

Perguntas frequentes

Why is long audio slow?: Inference runs on your device's CPU. Depending on the device, a few minutes of audio can take longer than real time. Choose the Tiny model when speed matters.
Which audio formats work?: Any format your browser can decode (WAV, MP3, M4A, WebM, Ogg, depending on the browser). If decoding fails, an error is shown.