Auscultation of heart sounds is an essential step for diagnosing heart diseases. Automatic auscultation using computer techniques is used to help physicians in their diagnosing process. Several researches are conducted for analyzing heart sounds using computer techniques. In this paper, new methods of extracting features from heart sounds are presented using Bark spectrogram cepstral coefficients (BSCC) and Mel-spectrogram cepstral coefficients (MSCC). Classification of normal and abnormal heart sounds are based on support vector machine or deep learning neural networks. Database of heart sounds are selected from Physionet challenge database. Signals of heart sounds are detrended and normalized. Then, wavelet transform is applied. After that, energy entropy is calculated. Then BSCC are considered. The classifiers of support vector machine (SVM) and deep learning (DL) with bi-long short term memory (BILSTM) are applied. The maximum obtained accuracy rate of 99.54% is achieved by using BSCC algorithm. The maximum obtained area under curve (AUC) of 0.9846 is achieved when using BSCC algorithm.