Sequential classification for articulation and Co-articulation classes of Al-Quran syllables pronunciations based on GMM-MLLR

Abstract

The overall study has been observed within the speech transformation of the Quranic recitation from the speech production to the measurable incorrectness of Al-Quran recitation pronunciations. The proposed methodology has used numerous techniques such as Mel Frequency Cepstral Coefficient (MFCC) features extraction, Gaussian Mix Model-Maximum Likelihood Linear Regression (GMM-MLLR) acoustic model, and sequences of model-based classification. Technically, the proposed acoustic hybrid features' properties of MFCC with derivatives MFCC (shift delta co-efficient (SDC) was fragmented into three-band frequency ranges. These are applied to the syllables of recitation speech which are represented with the combination of vowels and consonants characteristics. Each syllable is also classified based on the prolongation (Harakaat) types, syllables recitation rules (Tajweed) names, speech articulation (Makhraj) placement, and speech co-articulation (Makhraj sifaat) placement using three types of the conventional classifier which are Linear Discriminant (LD), Support Vector Machine (SVM) and K-Nearest Neighbour (KNN). Consequently, each of the conventional classifiers is applied in the form of sequential classification. The overall performance of band 1 has produced the results as 34.75%, 44.68%, and 92.24% for LD, SVM, and KNN respectively.