EVALUATION OF A PERIODIC/APERIODIC SPEECH DECOMPOSITION ALGORITHM

In this paper an evaluation method for periodic/aperiodic speech decomposition algorithms is proposed and the results of its application on a recently presented algorithm are discussed. The performance and the characteristics of the algorithm are examined both in the time domain, through HNR measurements and in the frequency domain, through spectral distance measurements criteria. As stimuli, synthetic steady-vowels distorted by natural-like sources of pseudoperiodic and noise excitation, with controlled varying time-frequency characteristics, were used. The analysis was used to test the influence of the fundamental frequency, perturbation type and perturbation level at the efficiency of the decomposition algorithm. The results demonstrated that the new algorithm is a powerful tool for analysing the aperiodic component of speech signals, under some conditions that have been explicited.