Unveiling Lombard speech flavors: A comprehensive categorization based on perceived intelligibility beyond loudness

Area of Science:

• Acoustic Phonetics
• Speech Communication
• Psychoacoustics

Background:

• The Lombard effect describes involuntary speech adjustments in noise to maintain intelligibility.
• Existing research often focuses on loudness changes, overlooking other acoustic modifications.

Purpose of the Study:

• To categorize Lombard speech into distinct "flavors" based on intelligibility beyond loudness.
• To investigate non-loudness acoustic modifications that enhance speech intelligibility.
• To identify specific acoustic features associated with different Lombard speech flavors.

Main Methods:

• Collected a Lombard speech dataset from 10 speakers across 11 noise levels (30-80 dBA) and two noise types (steady speech-shaped noise and dynamic babble).
• Utilized iterative paired-sample t tests to assess intelligibility of loudness-controlled Lombard speech.
• Performed acoustic analysis on identified Lombard flavors, examining fundamental frequency (F0), alpha ratio, formant frequencies, and vowel characteristics.

Main Results:

• Identified three Lombard flavors for steady speech-shaped noise (LF.0, LF.1, LF.2) and four for babble noise (LF.0, LF.1, LF.2, LF.3), with LF.0 being plain speech.
• Higher Lombard flavors showed significantly increased F0, alpha ratio, formant frequencies (F1/F2), vowel duration, and vowel ratio.
• These acoustic changes correlated with progressively higher perceived intelligibility.

Conclusions:

• Lombard speech intelligibility is enhanced by non-loudness acoustic modifications, categorized into distinct "flavors".
• Findings provide insights into intelligibility mechanisms beyond increased vocal effort.
• Results can inform the development of speech technologies like hearing aids and telecommunication systems to optimize speech clarity.