Speech Emotion Recognition (SER) is necessary for improving human-computer interaction to give systems the ability to identify and respond to speech emotions. This paper introduces an integrated deep learning model that combines LSTM with CNN for classifying emotions from speech data. The proposed approach uses a blend of temporal and spectral audio characteristics, including spectral contrast, chroma features, and MFCC (Mel-frequency cepstral coefficients), to enhance emotion recognition accuracy. CNN is employed to extract spatial dependencies from these features, while LSTM captures the sequential patterns present in signals for speech. The Ryerson Audio-Visual Database of Emotional Speech and Song dataset, also known as RAVDESS, is used to evaluate the model’s performance, consisting of samples of speech from eight different emotional categories. Experimental findings reveal that The accuracy of the hybrid CNN-LSTM model is of approximately 97.7%, surpassing conventional single-model approaches. This technique presents a reliable and effective solution for real-world emotion detection, contributing to advancements in affective computing and interactive AI systems.