Abdelaziz O., Kamel Y.

The evaluation of speech codecs can only be performed subjectively by listening tests. For practical reason, these listening tests cannot be performed in real time or in repetitive way. Several methods to evaluate the objective speech quality are available in the literature. The main purpose of these objective measures is to produce a more convenient alternative to the subjective tests that are expensive and time-consuming. Among these objective measures, one finds the traditional distortions as the SNR , the segmental SNR , the Itakura distance , and the cepstral distance . These measures of quality applied to low rate speech codecs produce an estimation that is not well correlated with the human perception. In the present paper, we describe a new objective measure to estimate the subjective quality obtained in the listening tests of speech codecs used in telephonic band (300-3400 Hz). This objective measure uses the recent time-frequency analysis called the Empirical Modes Decomposition (EMD), and estimates the amount of perceptible distortion by means of perceptuels models that we have adjust to the refined nature of Hilbert spectrum obtained by the EMD method. The obtained results indicate a satisfactory performance of our measure compared to the recent objectives measures, with possibilities of improvement by integrating other processes made possible by the use of the EMD analysis method. The nature of the EMD method permits us to take in consideration another types of distortions as distortions of short length and the local variations of the gain of the codec.