This paper employs the use of a Deep Recurrent Neural Network approach namely Long Short-Term Memory (LSTM) to predict gender and singer name by analyzing audio vocal portions. The ultimate aim of this paper is to build two Long Short-Term Memory (LSTM) models, one for predicting singer gender or gender identification and the other for classifying singer name. The accuracy of different existing algorithms such as SVM, CNN and MLP is then compared to the LSTM algorithm. The MIR-1K dataset that contains audio recordings from singers is used to train all algorithms, including LSTM, SVM, CNN and MLP, with LSTM being the proposed algorithm and SVM, CNN and MLP being existing algorithms, to perform this integration model. The proposed LSTM approach for a deep recurrent neural network offers better performance than other existing ones. The obtained results show that the effectiveness of the proposed model is used together with a good enough feature vector which works well than the existing methods.

Music information retrieval (MIR), Speaker identification (SPID), Deep learning, LSTM, RNN, MLP Network, SVM, Classification