Abstract
Sentence intelligibility in Japanese was tested in exponentially decaying or increasing echo and steady state disturbing noise conditions. The test items are sentences which are presented at condition of four echoes and in random noise that has spectrums simulated cheering yell. The noise and speech signals were presented to the listners from duplex speakers in an anechoic chamber. The following echo and noise conditions were used : echo numbers of 4 include the direct sound; echo durations of 60 to 190 msec ; echo decay rate of -6 to 6 dB per each echoes; noise levels of -10 to 0 dB RE speech level. Method of predicting sentence intelligibility in echoes is proposed, and evaluated in terms of the testing results. The method which proposed here assumes that the value of syllable articulation in a real field would be the same as what would be calculated were the disturbing echo energy replaced by equivalent noise. Three combinations of the following calculated syllable articulation in echoes were considered : Svv of vowel is interfered by echo vowels ; Scv of consonant is interfered by echo vowels; Scc of consonant is interfered by echo consonants. Modelled ten CV syllables, composed of two phrase and one pause, were used in this predicting method to calculate the articulation index. The articulation index computed by 1/1 octave band method of K. D. Kryter [J. Acous. Soc. Am. Vol.34, 1689-1697 (1962)]. Sentence intelligibility was obtained by multiple regression analysys between measured data and computed intelligibility which is converted from syllable artuculation by using the relation between articulation index and speech intelligibility. There were high correlation between the results obtained in anechoic chamber and proposed predicting method, 0.9853 of the coefficient of correlation was obtained by multiple regression analysys.
