The aim of the paper is to analyse the reliability of predicting human emotion from speech using Long Short-Term Memory (LSTM) approach. Human-computer interaction is extensively utilised in Speech Emotion Recognition (SER) to examine various methods and datasets for determining the practical results and discover more regarding this unsolved problem. While it is complicated to identify audio and difficult to estimate a person’s emotion because emotions are personal, SER makes this possible. Tone, pitch, expression, behaviour, and other states are used to determine emotion. A specific few of them are assumed to be capable of detecting emotion in speech. Here, different datasets are employed to recognize the emotions, such as Interactive Emotional Dyadic Motion Capture (IEMOCAP), Ryerson Audio-Visual Database of Emotional Speech and Berlin (EMO-DB). The final outcomes attained in trails, evidently demonstrate that the presented approach is used to achieve the task of voice emotion recognition. The proposed method implemented in python software for verifying its performance.