Audio classification is a difficult task because of the issue of extracting and choosing the optimum audio features. To reduce the computational complication in existing methods, this paper proposes a Feature Selection method based on Modified Bacterial Foraging Optimization Algorithm (MBFOA) for classification of audio signal. Enhanced Mel Frequency Cepstral Coefficient (EMFCC) and Enhanced Power Normalized Cepstral Coefficients (EPNCC) with peak and pitch are estimated as a signal feature and optimized using MBFOA with the fitness function. Using the Probabilistic Neural Network (PNN), audio signal is classified into music and speech signal. Then, if the signal is a music, the signal is classified as cello, clarinet, flute, etc. If the signal is detected as a speech, then it is again classified as a male or female voice. This approach shows that it is possible to boost the classification accuracy by using different features and optimization techniques.