Subjective test method for quantifying speaker identification accuracy of bandwidth-limited speech

Abstract

This paper proposes a subjective assessment test procedure that quantifies speaker identification accuracy of bandwidth-limited speech. To quantify such accuracy, it is common to conduct a subjective assessment test and regard the correct answer rate as speaker identification accuracy. In such a test, the following two requirements should be met. (1) Conduct an experiment in which stable assessment is executed over time and (2) obtain an evaluation value that has little dependence on the degree of identification difficulty of the unprocessed speech used in the test as speaker identification accuracy. Our proposed subjective assessment test procedure takes the above requirements into account. We show that the proposed test procedure is effective in quantifying speaker identification accuracy of bandwidth-limited speech by verifying the validity of this procedure from obtained test results.