Speech recognition (ASR, speech-to-text)

Speech recognition, also called ASR or speech-to-text, turns spoken audio into written text.

  • Models and architecture
  • Updated

An ASR model listens to a recording or a live stream and writes a transcript, often with timestamps and punctuation and, in some models, speaker labels. Some models transcribe finished files, while streaming models output text as you speak and let you trade a short delay against accuracy. Many also translate or transcribe dozens of languages.

Accuracy is usually reported as word error rate on public test sets that rarely match your audio, so accents, background noise and domain words can change results a lot. Whisper can also write text that was never spoken, so check the transcripts that matter.

Related terms

  • Word error rate (WER) Word error rate is the standard score for speech recognition: the share of words in a transcript that are wrong, missing or extra compared with a correct reference.
  • Speaker diarization (speaker labels) Diarization works out who spoke when in a recording and labels each stretch of speech with a speaker, such as Speaker 1 and Speaker 2.
  • Text-to-speech (TTS) Text-to-speech models turn written text into spoken audio, and many can copy a voice from a short sample.

Keep reading

All terms