Definition
Automatic speech recognition (ASR) is the technology that transcribes spoken audio into text automatically, also commonly called speech-to-text (STT).
How it works
Earlier systems modeled acoustics and language separately, while modern end-to-end models map audio features directly to text sequences, coping better with casual speech, accents and noise. Word error rate (WER) is the standard accuracy metric.
Usage
Meeting and interview transcription, video captioning, voice keyboards, voice assistants and customer-service analytics all start with an ASR system.
Related and confusable terms
Text-to-speech (TTS) runs in the opposite direction: ASR turns speech into text, TTS turns text into speech, and the two are often paired in spoken dialogue systems.