Generative AI & the senses
Speech recognition is the technology that converts spoken words into written text, powering dictation, voice assistants, and automatic captions.
Sound arrives as a wave of pressure changes. Speech recognition turns that wave into the words that were said. It has to cope with accents, background noise, fast talkers, and words that sound alike. Modern systems do it with neural networks trained on many thousands of hours of transcribed audio.
Older systems worked in stages: detect sounds, guess syllables, then guess words using a dictionary. Today's models often go straight from audio to text in one network, and some handle dozens of languages. Accuracy on clear speech is now close to that of human transcribers.
Speech recognition is only the listening part. Understanding what you meant, and replying, are handled by language models. Its cousin, text-to-speech, goes the other way and produces spoken audio, which is how assistants talk back.
You dictate a text message while walking and your phone types it out, correctly choosing 'their' over 'there' from the surrounding words.