Speaker recognition helps determine who is speaking by using the speaker’s unique voice characteristics. In this paper, we have implemented a Matlab code using SincNet Neural Architecture with the standard CNN. We aim to show that SincNet will have accuracy better than CNN as it is customized to learn the required characteristics of the voice. We create a model and train it on the dataset which contains only continuous speech of same language audio files. While testing, we will test the model by giving different languages read audio files of the same speaker. By this, the model becomes more robust and reliable. In read audio, the speaker is given a paragraph to read, while continuous/spontaneous audio is a normal conversation speech. In read speech, the person is generally aware of the content thus, the flow of words is controlled. Spontaneous speech is raw and unfiltered. Speakers respond immediately depending on their thoughts. The speaking style of the same speaker can be different for different speakers. It also depends on their emotional state. The paper also shows how speaker verification varies when different styles of speaking are employed. Taking into consideration all the factors, the proposed model was trained and validated.