Sound becomes text. Speech recognition transcribes spoken language into written form, handling accents, background noise, and continuous speech. It powers voice assistants, dictation tools, and automated call centres.

Classical systems used hidden Markov models with handcrafted acoustic features. Modern systems use neural networks trained end-to-end on paired audio and text, learning acoustic and language patterns together.

Common speech recognition tasks

• Dictation and transcription
• Voice commands and assistants
• Real-time captioning
• Call centre analytics
• Meeting summarization

Accuracy is high for clear speech in quiet conditions. It drops with heavy accents, overlapping speakers, or noisy environments. Word error rates on benchmarks do not always predict performance in the field.