Multi-objective Approach to Speech Enhancement Using Tunable Q-Factor-based Wavelet Transform and ANN Techniques

The tunable Q-factor-based wavelet transform (TQWT) is a novel method employed for the speech enhancement (SE) task. However, in TQWT, the controlling parameters Q-factor and the level of decomposition (J) are kept constant for different noise conditions which deteriorates the overall performance of SE. Generally, the performance of SE is calculated in terms of quality and intelligibility. However, it has been reported that these two evaluation parameters do not always correlate with each other because of the distortions introduced by the SE algorithms. These two important issues are addressed in this paper, and satisfactory solutions are provided by employing a multi-objective formulation to find the optimal values of the Q and J of the TQWT algorithm at different noise levels. In addition, to correctly estimate the appropriate values of Q and J from the unknown noisy speech, a low complexity functional link artificial neural network-based model is developed in this paper. To assess the performance of the proposed hybrid approach, subjective and objective evaluation tests are carried out using three standard noisy speech data sets. The results of the study are computed with six recently reported SE methods. It is demonstrated that in both the subjective and objective evaluation tests, the proposed hybrid approach outperforms the other six SE methods.