Influence of sound source characteristics in determining objective speech intelligibility metrics

Sound source characteristics may be one of the main causes of objective speech intelligibility metric inaccuracy. In this study, the influences of the sound source directivity and frequency response were investigated using three typical sound sources: an artificial mouth, a monitor speaker, and a dodecahedral sound source. The results show that, the simultaneous influences of directivity and frequency response on the objective speech intelligibility metric are significant, typically with a variation of 0.147 in speech transmission index (STI); sound source directivity may also result in a noticeable difference in the objective speech intelligibility metric, typically with a variation of 0.123 in STI. In comparison with sound sources with a high directivity index (DI), the measurement results for sound sources with a relatively low DI may be higher when background noise is high, and may be lower when background noise is low. The influence of sound source directivity may also depend on the room acoustic conditions, and at receiver position where reflections are abundant, the influence of sound source directivity may be more significant. Not applying frequency response equalisation resulted in large errors in the values being measured, which deviate from the real values of STI by up to 0.172, depending on the original frequency response characteristics of the sound sources that are used.