Peer-reviewed research
An Audio-based Sequential Punctuation Model for ASR and its Effect on Human Readability
Gyorgy Szaszak, Acta Polytechnica Hungarica, 2019
Read the originalRead on October 16, 2026
- What it is
- A 2019 journal paper on inserting punctuation into speech recognition output with a prosody-based model. It argues that false punctuation marks disturb readers more than missing ones, and it describes dictation systems as built around explicit spoken punctuation.
- How much weight it carries
- Peer-reviewed journal article with listening tests on Hungarian read speech. The findings concern one language and one model; the precision argument is the authors' reasoning, backed by a perception test rather than a measured typing task.
Facts taken from this source
- The 2019 paper says that for dictation systems, where punctuation matters most, a telegraphic style with explicit dictation of the punctuation marks was foreseen.
“a telegraphic style alike explicit dictation of the punctuation marks was foreseen” (Section 1, Introduction)
- The 2019 paper argues that automatic punctuation should favor precision over recall, because readers can add a missing mark in their heads more easily than remove a wrong one.
“We suppose that precision is more crucial in this sense than recall, i.e. false detections should be minimized, and human reader should rather be required to insert missing punctuations ... than eliminating incorrect ones” (Section 1, Introduction)
- The 2019 paper proposes and tests a punctuation model built on prosody, the pitch and timing cues of speech, rather than on the words alone.
“proposes and evaluates a prosody inspired approach and a phrase sequence model” (Abstract)