Abstract For over a century, researchers and scientists worldwide have been fascinated by electrical signals since the first electrocardiogram was produced. These signals have been the focus of extensive research and development efforts, leading to remarkable advancements in cardiology. It is widely known that our mental states significantly impact the activity of our Autonomous Nervous Systems, which in turn affects the signals of our electrocardiograms. This has led to the exploration of methods for identifying emotional states using deep learning algorithms to reach high degree of emotional identification accuracy. The Dreamer emotional recognition dataset is used to train the model. Due to the limitation of the dataset, a new method of dataset pre-processing is developed using four distinct lowpass filters for the purpose of data augmentation to the electrocardiogram signal. A hybrid architecture of Bidirectional Long Short-Term Memory and Convolutional Neural Network is used to detect nine emotions (amusement, excitement, happiness, peacefulness, anger, disgust, fear, and surprise). The proposed thesis delivers an impressive accuracy rate of 98.6%. Compared to other studies, this level of precision is the highest achievable accuracy. Compared to other studies, this level of precision is the highest possible accuracy. This thesis demonstrates the potential of electrocardiogram signals for detecting emotional states. Furthermore, it highlights the importance of comprehensive pre-processing techniques and data augmentation to enhance the accuracy and precision of the models. The findings of this thesis provide a valuable foundation for future research in this area, with potential applications in fields such as mental health, psychology, and human-computer interaction.