An ensemble learning model for single-ended speech quality assessment using multiple-level signal decomposition method

In this study a novel process for measuring quality of speech is introduced which employs multiple-level decomposition of signal method. In this way, the Discrete Wavelet Transform (DWT) is used to decompose original speech signal into different frequency sub-band signals. Then the feature vectors are obtained by extracting MFCC features from each sub-band. In order to investigate the capabilities of ensemble learning methods, various ensemble regression models are studied and results are compared with individual models. Also, to prepare training and test dataset, a simulation environment is set up which distort speech signal by different speech impairments. At last, different experiments are performed to illustrate the efficiency of ensemble methods. Results demonstrate that using a group of base learners (ensemble model) improve the performance of models in comparison with single learner.