Detection of gender from voice samples finds potential application in domains where it is required to analyze and interpret human interactions and emotions, recognize speakers etc. It is especially useful for human computer interaction systems, biometric identification, targeted marketing, speaker recognition systems, psychology and social science research. It can be considered as a signal processing problem where speech features in the first stage are extracted and explored to select the potential ones on which suitable machine learning (ML) and deep learning (DL) techniques are applied in the second stage. In the present work, our focus is on extraction of two frequency domain features namely pitch and Mel Frequency Cepstral Coefficients (MFCC) for identification of gender from voice samples. The datasets used for training and testing of ML and DL are namely RAVDESS and Mozilla Common Voice dataset that contain recorded voice samples of both genders, i.e., male and female, from a range of sources. Five popular ML classifier models are subjected to training and testing on the features extracted from the preprocessed voice samples from the datasets. The chosen models are Support vector machine (SVM), Logistic Regression (LR), Decision Tree (DT), Random Forest (RF) and Naïve Bayes’. Accuracy, precision, recall, F1 score and average prediction time are the performance measures chosen for comparing the performances of ML and DL models. Detection accuracies obtained from ML models are quite high and differ from dataset to dataset. With a view to boosting the performance further, a convolutional neural network (CNN) model is proposed for accomplishing the same task. The accuracy obtained from the proposed CNN model is higher than those obtained from ML models. The preliminary investigation results are promising and vital for theoretical extension and real time implementation.