Automatic Speech Recognition
Also known as ASR
Overview
Automatic speech recognition (ASR) studies how to turn audio into written words, supporting tasks such as transcription, captioning, voice interfaces, and clinical documentation. Systems commonly use self-supervised speech encoders with CTC or transducer decoders, encoder-decoder models, and increasingly speech-language models; adaptation, contextual biasing, and synthetic or unlabeled data help handle specialized vocabulary and underrepresented languages. Work also targets streaming and long-form audio, overlapping speakers, noisy recordings, and audiovisual cues, with evaluation on datasets such as LibriSpeech, Common Voice, and meeting and domain-specific benchmarks. Strong pretrained models coexist with persistent gaps in low-resource and atypical speech, dialect fairness, hallucination control, and real-world robustness, while word error rate alone often misses meaning, usability, and downstream impact.
Latest papers
484 papers67 papers in the last four weeks, down 7% on the four weeks before. 0.4% of all new papers.