Evaluation of single-channel speech enhancement algorithms by using objective quality and intelligibility measures

Abstract

In this study, single-channel speech enhancement algorithms were evaluated with objective quality and objective intelligibility measures using Turkish speech database. The clean 30 sentences from the METU database are corrupted by car and babble noise types at −10, −5, 0, 5 and 10 dB SNR levels. The Karhunen-Loeve Transform has been found to be more successful than other methods in terms of both quality and intelligibility, given the amount of segmental SNR improvement, weighted spectral slope, short-time objective intelligibility values and spectrogram representations.