The proposal of quantification method of speaker identification accuracy for speech communication service

摘要

We propose a method for quantifying how much the signal processing in speech communication services affects speaker identification accuracy. In this quantification using a subjective speaker identification test, the following four requirements should be satisfied, (a) The results should be less dependent on the order of speech samples, (b) Various speakers should be used in the test, (c) Speech samples with uncommon characteristic unrelated to voice quality should be excluded, (d) Speech samples that are difficult for subjects to identify the speaker even in the best speech quality should be excluded. We propose a method of testing subjective speaker identification to satisfy them. By using this method, we quantify the effects of bandwidth limiting speaker identification accuracy.