Whisper

Turning spoken audio into written text, in many languages, is what Whisper does. Whisper is an open source speech recognition model from OpenAI, published on GitHub, and you can use it as an AI transcriber or speech-to-text engine.

It also translates speech and identifies which language is being spoken.

Under the hood, Whisper is a Transformer sequence-to-sequence model trained on a large, varied set of audio.

Tasks like multilingual recognition, speech translation, language identification and voice activity detection are all expressed as tokens the decoder predicts, so one model covers steps that older speech pipelines split across several components.

Special tokens tell the model which task to perform.

Because it is a model rather than a finished app, expect to run it yourself or build it into your own software.