The effects of frequency lowering and spectral degradation on low pass filtered speech

Abstract

Nonlinear frequency compression has shown potential benefits for improving speech understanding for people with high frequency losses. This study examined the effect of changing start frequency when nonlinear frequency compression is applied at a very low frequency range. Another important area of research is in understanding the role of different acoustic features to speech understanding. Two such features are envelope (and its complement temporal fine structure) and fundamental speech frequency (F0) (and its complementary harmonics). This study examined the effect of envelope removal and fundamental frequency flattening on recognition of low pass filtered speech, and of possible interactions with frequency lowering. To investigate these two questions, 18 normal-hearing subjects listened to Mandarin speech processed using different combinations of low pass filtering, nonlinear frequency compression with a varying start frequency, and envelope removal or F0 flattening. Results showed that lowering start frequency to 200 Hz did not significantly affect performance, but worsened performance with a start frequency below 200 Hz. This can be attributed to the fact that the region between 0 Hz and 200 Hz contains the fundamental frequency. Envelope removal and fundamental frequency flattening both worsened performance, indicating that low pass filtering removes some of the redundancy contained in temporal fine structure and also removes the harmonics of fundamental frequency that normally preserve performance in full frequency spectrum speech. Fundamental frequency flattening worsens performance more than envelope removal, showing the importance of preserving fundamental frequency contour for better understanding of frequency-compressed speech. Finally, the benefits of nonlinear frequency compression disappear when envelope is removed, showing that an intact envelope also contributes to the intelligibility advantage of frequency-compressed speech.