Perutterance

Perutterance is a term used in certain technical contexts to denote analyses conducted on a per-utterance basis in a spoken dialogue or audio stream. It is formed from per and utterance and is often seen as per-utterance, per utterance, or simply perutterance in software documentation or some research papers. The concept describes treating each utterance—the speech segment bounded by turn boundaries or silences—as the basic unit of analysis, rather than processing entire recordings or aggregating data across multiple speakers.

Common applications include transcription evaluation, where perutterance metrics, such as perutterance word error rate, are computed for each utterance; prosodic or acoustic analysis performed for every utterance; and diarization or speaker-change evaluation measured per utterance. This approach provides fine-grained diagnostics, enabling the identification of segments that degrade overall performance, such as brief, unclear, or overlapped utterances.

Terminology and standardization: There is no universal definition or standard metric for perutterance; authors may differ on whether the unit includes punctuation, whether partial utterances are counted, or how to handle partial transcriptions. As a result, when encountered, it is advisable to consult the specific source for exact scope and calculation.

Relation to related terms: Perutterance is related to general per-utterance analysis in speech processing; the hyphenated form per-utterance is more common in formal writing, while perutterance appears in code comments or system labels.

See also: utterance, per-utterance, word error rate, diarization, prosody.