Recognition of Mandarin words heard through single and multiple spectral bands

Abstract

Two listening experiments were designed to evaluate the applicability of two macroscopic models—the articulation index and the speech recognition sensitivity (SRS) model—towards Mandarin speech recognition. The experiments aimed to identify if there is any synergetic and redundant effect of spectral frequency bands in monosyllabic Mandarin words. In experiment 1, 20 normal-hearing subjects listened to monosyllables through 20 bandpass filters, which have center frequency ranging from 125 to 8000 Hz, 1/3-octave bandwidth, and filter slope of 100 dB/octave. The purpose of experiment 1 was to identify the intelligibility of spectral frequency bands in Mandarin speech. In experiment 2, 10 subjects listened to monosyllables with the combination of 2 or 3 individual bands for investigating the synergetic or redundant interactions. The perceptual outcomes of the second experiment were fitted to both models by statistical regressions, and the results indicated that the contribution of each band to recognition of Mandarin monosyllables was not independent, and the recognition of Mandarin tones, consonants, and vowels showed different redundant and synergetic effects. The SRS model appears to be able to predict the differential effect of linguistic entropy on filter condition.