In dealing with continuous Chinese speech

this paper proposes a novel method (RBFOBSQ)

using Radial Basis Function Neural Networks (RBFNNs)

for output-based objective speech quality assessment.First the characteristics of speech signals at the output of the speech system were extracted by Mel cepstrum coefficients.Then the mapping from the characteristics to the Mean Opinion Score (MOS) was accomplished by the RBFNNs

and the average value of the RBFNNs' outputs was the result of output-based objective speech quality.The experimental results show that the correlation degree reaches more than 0.92 when using the train samples and attains more than 0.88 when using the test samples.