Automatic speech recognition (ASR)

Automatic speech recognition (ASR) is the technology that converts spoken language in audio into machine-readable text, powering transcription, captions, dictation and voice assistants.

Modern ASR systems are end-to-end neural networks — models like OpenAI's Whisper are trained on hundreds of thousands of hours of multilingual audio and output punctuated text directly, replacing the older pipeline of separate acoustic and language models.

ASR quality varies with audio clarity, accent coverage, vocabulary and language resources: a clean podcast in English transcribes near-perfectly, while noisy phone audio in a low-resource language still deserves a human review pass.

Related terms

Word error rate (WER)Word error rate (WER) is the standard accuracy metric for speech recognition: the number of word substitutions… Acoustic modelAn acoustic model is the component of a speech-recognition system that maps audio features to the sounds of sp… Language model (in ASR)In speech recognition, a language model supplies knowledge of which word sequences are likely, letting the sys… End-to-end speech recognitionEnd-to-end speech recognition uses a single neural network that maps audio directly to text, replacing the cla… Streaming speech recognitionStreaming speech recognition transcribes audio as it arrives, emitting words within moments of them being spok…

Put the term to work

Transcribe audio or video in 99 languages — speaker labels, word timestamps, captions.

Transcribe a file →