Hall-specific cepstral correlates of Mandarin monosyllable recognition in legacy room-acoustic recordings

This study reanalyzes legacy Mandarin monosyllable recordings obtained in seven room-acoustic environments using a differential Mel-frequency cepstral approach (ΔMFCC). Cepstral deviations were calculated using the corrected cepstral-axis mapping over the early F2–F13 region and averaged across 36 syllables at each receiver position. Speech recognition scores (SRS), Speech Transmission Index (STI), signal-to-noise ratio (SNR), and room-transformed recordings were used to evaluate whether ΔMFCC provides information complementary to STI. For each hall, dCep2–dCep13 were examined over a hall-specific retained receiver subset to identify the cepstral order most strongly associated with SRS. The dominant dCep differed across halls, demonstrating that the relevant cepstral response was space dependent rather than universal. Within halls, the selected ΔMFCC measure showed strong negative associations with SRS, whereas STI exhibited a more coherent trend across halls. Exploratory cross-hall analysis further indicated an association between hall-level SNR and the signed ΔMFCC–SRS regression slope. These findings support ΔMFCC as a complementary, hall-specific cepstral descriptor of room-induced speech degradation, while STI remains the more suitable standardized reference for comparisons across venues.