Segmental durations in connected speech signals: preliminary results

Area of Science:

• Phonetics and Speech Science
• Computational Linguistics
• Acoustic Phonetics

Background:

• Understanding speech timing is crucial for phonetics and speech technology.
• Previous research often focused on isolated sounds rather than connected speech.
• Variability in speech rate (fast vs. slow talkers) presents challenges in acoustic analysis.

Purpose of the Study:

• To create a database and methodology for studying phonetic unit durations in connected speech.
• To compare durational characteristics between fast and slow talkers.
• To investigate contextual influences on phonetic durations in informal speech.

Main Methods:

• Selection of talkers into fast and slow groups based on reading time.
• Segmentation of recordings into breath groups, pauses, and phones using computer graphics and digital playback.
• Identification and measurement of stop consonant release portions.
• Calculation of first-order statistics and comparison with existing literature.

Main Results:

• Analysis revealed differences in breath groups, phonation time, and pauses between fast and slow talkers, but no script-independent predictive measures.
• Mean durations of phonetic categories showed similar percentage changes across talker groups.
• Contextual influences on durations showed expected changes, with some deviations from traditional predictions in informal speech.
• Less than 50% of stop consonants exhibited measurable releases.

Conclusions:

• While distinct patterns exist between fast and slow talkers, individual speech rate is difficult to predict from these measures.
• The study provides a valuable database and methodology for further research into connected speech timing.
• Findings suggest that informal connected speech may exhibit durational patterns not fully captured by traditional phonetic models.