Transforme gravações, áudios de voz e vídeos em texto ou legendas com a IA Whisper, em cerca de 100 idiomas.
Speech recognition by OpenAI Whisper (MIT License), running with Transformers.js.
No. The speech recognition model (OpenAI Whisper) is downloaded to your browser once and runs there, so your recording never leaves your device. That also means it works with confidential meetings and interviews.
Fast (42 MB) works well for clear speech in English and other widely spoken languages. Accurate (77 MB) makes fewer mistakes, especially with accents, background noise and other languages, but takes about three times as long. Both are downloaded once and then cached.
It depends on your device. On a recent laptop, the Fast model transcribes a minute of speech in roughly 10 to 30 seconds. Phones are slower. Keep the tab open until it finishes.
Whisper understands about 100 languages and can also translate speech in any of them into English. Accuracy is best for English, Spanish, French, German, Portuguese, Italian, Japanese and Chinese. Less common languages, including Sinhala and Tamil, are recognized less reliably by these small models.
Choose your video, turn on timestamps, and download the SRT or VTT file when it's done. Most video players, YouTube and editing apps can load these subtitle files.
