Definition
VideoText editorial team. Updated September 13, 2026.
Automatic speech recognition (ASR) is software that maps an audio waveform to text without a human typing every word. Modern ASR systems are statistical or neural models trained on large collections of speech and transcripts.
The sections below explain how automatic speech recognition is used in transcription and subtitle work, what people often mix up, and which VideoText workflow applies when the next step is a file or review pass.
An ASR system converts sound into features, then predicts a sequence of tokens or words. Training data and decoding rules determine how it handles accents, noise, and punctuation.
The 2022 Whisper paper showed that a model trained on 680,000 hours of weakly labeled web audio can be evaluated zero-shot on public test sets. That is a research result, not a promise about every real recording.
ASR makes it practical to draft transcripts and captions for long media. It does not, by itself, guarantee an accessible caption file. Organizations in 3Play’s 2024 survey still treated unedited automatic captions as insufficient.
The same Whisper Large V2 checkpoint scored 2.7% WER on LibriSpeech Clean and 25.5% WER on CHiME-6 in the authors’ table — one model, two very different conditions.
“ASR accuracy” is not a single number. Always ask: which dataset, which language, which microphone, how many speakers, and was the text normalized?
