Noise Handling Techniques in DBSCAN Clustering for Adult and Child Voice Diarization

Speaker recognition technology plays an important role in voice processing and security, especially in authentication and security monitoring. Based on DBSCAN clustering algorithm, this study explores its application in multi-speaker speech data. This paper introduces the background and significance of speaker recognition and applies the current common feature extraction method and clustering algorithm. Subsequently, we describe in detail the experimental design and data processing steps, including the acquisition and preprocessing of speech datasets, and feature extraction methods based on MFCC features. By analyzing and comparing the experimental data, we evaluate the experimental results of DBSCAN algorithm on clustering speakers under different parameter settings. The experimental results show that the DBSCAN algorithm can effectively distinguish speech segments of different speakers and demonstrate its robustness and applicability in complex environments. In summary, this study provides theoretical and empirical support for further exploration and optimization of DBSCAN-based speaker recognition techniques.