Speaker Diarization divides an audio recording into segments corresponding to individual speakers, which is pivotal in advancing speech-related applications. This paper introduces a novel approach to speaker diarization leveraging Long Short Term Memory (LSTM) networks, a recurrent neural network type known for capturing long-range dependencies in sequential data. The model was trained using a Recurrent neural network, one of the best neural networks for speech recognition. An alternate way of training the model was the Convolution neural network, which is good enough to recognize the images of the audio file as spectral. However, the proposed methodology uses RNN-LSTM and merges the capabilities of feature extraction techniques with that of deep learning model with the understanding of temporal design offered by LSTMs.