In this paper, we present a comparative study for gender and age recognition using speech signals. In this regard, we explored state-of-the-art deep neural networks to classify gender from speech features. We utilized two types of features, (i) time-series related features such as pitch, tone, intonation, etc., and (ii) transform speech features to spectrograms (using STFT and other techniques). We further evaluated the efficacy of each feature and deploy on deep learning algorithm (Resnet50). We also applied transfer learning for checking the accuracy of the dataset. The dataset utilized in our study is well-known speech dataset for gender recognition such as Common-Voice. We presented extensive results comparison with Sota techniques.