Determination of correlations of objective speech intelligibility descriptors in university classrooms considering the effect of air conditioning noise

Abstract

Speech intelligibility in classrooms is shaped by the interaction of direct sound, reverberant sound, and background noise level (BNL). In classrooms, air conditioners (ACs) are often switched on and run at different fan levels, dynamically changing the BNL and affecting speech intelligibility. Objective descriptors used to evaluate speech intelligibility are articulation loss of consonants (ALC), speech transmission index (STI), and useful-to-detrimental ratio (U50). This study examined the effect of ACs on speech intelligibility through calibrated SNR measurements and IEC-compliant STI analysis in five different unoccupied university classrooms. Measurements were calibrated using reference standards, and STI values were corrected according to IEC 60268-16:2020 Annex M requirements when signal-to-noise ratios fell below 20 dB. The results demonstrate that background noise level has a stronger influence on speech intelligibility than reverberation time, with noise control taking precedence over reverberation management in typical classroom environments. Contrary to literature claims, no significant correlation was found between STI and U50 (r = 0.029, p > 0.05), challenging the interchangeable use of these descriptors in HVAC-affected spaces. Statistical modelling revealed that BNL explains 92.1% of STI variance compared to 81.4% for reverberation time (T30), with micro-perforated panels improving intelligibility primarily through noise reduction rather than reverberation control. The correlations between STI and T30 weaken substantially with increased AC noise, particularly at mid-frequencies, while high-frequency relationships (2000–4000 Hz) remain most predictive of speech intelligibility. These findings underscore the need for case-specific acoustic validation over universal correlation assumptions in real-world classroom environments.