Gender Recognition Method from Speech Data and Analysis of the Developed Method in Emotional Categories

Recent advances in speech-based artificial intelligence have increased interest in extracting demographic information, such as gender, directly from speech signals. However, most of the existing studies have not addressed in detail the relationship between the emotional content of speech and gender prediction performance. This situation has created a gap in the literature. In this study, a deep learning-based gender recognition method was developed to fill this gap, and the results of the obtained model were analyzed in three different categories: polarity (positive, negative, neutral), emotion type (joy, fear, surprise, sadness, disgust, anger, neutral), and emotional intensity (normal, strong). The gender recognition performance was 96.18% in terms of accuracy. In addition to high performance, the model's analysis results show that gender prediction accuracy varies depending on the emotional content and intensity of the speech.