Multi Speaker Activity Detection Using Spectral Centroids

In recent years, there has been a notable increase in the demand for reliable and effective methods in the domain of audio processing, specifically for the purpose of analysing complex acoustic surroundings. Multi-Speaker Activity Detection (MSAD) is a fundamental task in this domain that entails the detection and classification of multiple speakers in an audio recording. In this work, we propose a novel approach for MSAD based on the concept of Spectral Centroids. Spectral Centroids provide valuable insights into the spectral characteristics of audio signals and have been widely used for tasks such as music genre classification and instrument recognition. In our work, we utilise spectral centroid data to create an efficient MSAD system. The performance of our proposed multi-speaker auditory Detection (MSAD) system is assessed using a broad dataset that encompasses a range of auditory situations and speaker configurations. The utilisation of gender-specific threshold values and the inclusion of the spectral centroid characteristic in the process of multi-speaker activity detection exhibit potential in effectively ascertaining the number of speakers in simultaneous voice recordings. The results obtained for male and female test sets indicate the effectiveness of this approach in differentiating between speakers of different genders. Continued research and refinement of the spectral centroid-based method could lead to advancements in multi-speaker activity detection systems, enabling applications such as speaker diarization and speech separation in complex audio environments.