Chant is a melody which is exist in church with an Ethiopian Christian religion. It is one part of song in Ethiopian orthodox church which is the most ancient spiritual practices and part of most religions and spiritual paths. Due to this, scholars or priests had used different accompanying instruments and features of song for better expression of their feeling. Drums and cymbals are instruments which used in Ethiopian Orthodox church for the purpose of hymn with different chanting levels. There are ten type of chants which classified based on their own properties. We considered only the five types of chants which songs with drums and cymbals instrument in prediction which are song frequently, but the chants which are song without drums and cymbals are not song frequently that are not available. Regarding to this we collected an audio chant sound data from Bahir dar Orthodox churches like S.t Giworgis cathedral church, Abune Gebere Menfes kidus church, Bezawit S.t mariam church, Kidane mihret church, S.t gebreal church on Sunday and feast days in which for kum 102, mereged 84, niuse mereged 120, tsifat 123 and wereb 126 using voice recorder and audio recorder tools. Then after we applied a signal processing approach in MATLAB software application and preprocessed the collected signal using techniques like silence removal, noise removal, framing, preemphasis and windowing, and feature extracted techniques like MFCC, LPCC and combined features using GMM algorithm. In addition, we used implementation techniques such as Audacity, audio converter and audio cutter. To evaluate the prediction accuracy, from the total of 555 data sets 80% were used for training and the remaining 20% was used for testing. We had compared the prediction feature vectors of MFCC, LPCC and combination of two with GMM and we predicted a well-known Chant levels like kum, Mereged, nius Mereged, tsifat and wereb which are song based on drums and cymbals instrumental sounds by combining MFCC and LPCC feature vectors with better result performance. The experimental result for kum, mereged, niuse mereged, tsifat and wereb is 76.1%, 81.9% ,83.3% ,84.7% and 85.8% respectively which were obtained using combined features. The overall prediction accuracy was 82.4%