A subjective evaluation of the effects of speech coding on the perception of emotions

The accuracy of the reproduction of emotions by speech coders has only recently been identified as a relevant issue. Several published studies have shown that speech compression reduces the accuracy of emotions classification. These studies, however, were all conducted using objective evaluation methods that involve an automatic classifier. The only definitive way to prove or disprove that the emotional content of a speech signal is degraded by compression operations is by testing it with human subjects. This paper proposes a subjective evaluation method and applies it to emotional speech coded by the AMR-WB speech coder at 6.6 and 12.65 kbps. The results confirm that there is a significant degradation in the perception of emotions by human listeners at both bit rates. The proposed evaluation method, and the insight provided by the results, could be useful in developing new speech coders that better preserve the emotional content of speech signals.