Glossary → Transcription & speech recognition
Transcription & speech recognition
Hallucination (in transcription)
In speech recognition, hallucination is when the model outputs fluent text that was never spoken — typically repeated phrases or invented sentences appearing during silence, music or noise.
Hallucination is a known failure mode of large speech models like Whisper, concentrated in long non-speech stretches. Pipelines defend against it with voice activity detection, silence trimming and repetition filters.
If a transcript contains a suspiciously fluent sentence during what you remember as silence, that is what you are looking at — spot-check against the audio.
Related terms
Voice activity detection (VAD)Voice activity detection (VAD) is the technique of finding which parts of an audio signal contain speech and w…
Automatic speech recognition (ASR)Automatic speech recognition (ASR) is the technology that converts spoken language in audio into machine-reada…
Confidence scoreA confidence score is the speech recognizer's own estimate of how likely each word or segment is correct, typi…
Put the term to work
Transcribe audio or video in 99 languages — speaker labels, word timestamps, captions.
Transcribe a file →