USR PUBLISHER

Speech Emotion Recognition (SER) has attracted considerable attention because it enables intelligent systems to recognize human emotions from speech, thereby improving the quality of human–computer interaction. Applications such as virtual assistants, healthcare monitoring, intelligent tutoring, and customer support increasingly rely on accurate emotion recognition. Nevertheless, reliable classification remains difficult because speech signals exhibit substantial variation across speakers, languages, recording environments, and emotional expressions. This study presents a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) model that jointly learns spectral and temporal information from speech signals. Mel-frequency cepstral coefficients (MFCCs) and Mel-spectrogram representations are used to describe the acoustic characteristics of speech, while the CNN extracts discriminative local patterns and the LSTM models temporal dependencies between successive speech frames. An end-to-end processing pipeline consisting of data preprocessing, feature extraction, model training, hyperparameter optimization, and performance evaluation is adopted to develop the proposed framework. The model is evaluated using benchmark speech emotion datasets and assessed with standard classification metrics, including accuracy, precision, recall, and F1-score. The experimental results indicate that combining convolutional and recurrent learning improves emotion recognition performance compared with individual CNN and LSTM models. The hybrid architecture also demonstrates improved robustness across multiple emotional categories and varying speech conditions. The proposed framework provides an effective and scalable solution for speech emotion recognition and can support applications in affective computing, healthcare, intelligent conversational agents, and Internet of Things (IoT) environments. Future work may extend the framework through attention mechanisms, multimodal learning, and lightweight architectures suitable for real-time deployment.