In recent years, recognizing genre of music involves in automatically assigning a label to a musical piece based on the genre. Traditional approaches for audio signal classification and music genre classification had faced several challenges which include poor temporal modeling, limited data representation and feature extraction limitations. Therefore, this research proposes Convolutional Recurrent Neural Network (CRNN) for audio signal classification for music genre recognition. Initially, data is collected from GTZAN dataset which consists of 1000 audio tracks. Then, this dataset is preprocessed by using normalization, and features are extracted by utilizing Mel-Frequency Cepstral Coefficients (MFCC) which transform audio signals into spectral representations by representing important features of signal. Next, features are selected by using mutual information (MI), where features are chosen based on mutual information with target variable. Finally, classification is done by using proposed CRNN and to improve its performance hyperparameter tuning is done by using Bayesian optimization with Tree-structured Parzen Estimator (TPE). The proposed CRNN achieved better results in terms of Accuracy (0.98), Precision (0.95), and F1-Score (0.97) when compared with existing Modified Convolutional Neural Network (Modified CNN).