Automatic Speech Recognition and Translation for Low Resource Languages

19Comprehensive Analysis of State-of-the-Art Approaches for Speaker Diarization

Trisiladevi C. Nagavi*, Samanvitha S., Shreya Sudhanva, Sukirth Shivakumar and Vibha Hullur

S. J. College of Engineering, JSS Science and Technology University, Mysore Karnataka, India

Abstract

Speaker diarization is the ability to compare, recognize, comprehend, and segregate different sound waves on the basis of the identity of the speaker. As an illustration of this theory, different ways to achieve these objectives are analyzed in this book chapter. Speaker diarization can prove to be crucial in the future with regards to the field of education, healthcare, forensics, smart traffic management, media, etc. There are numerous steps associated in the process of speaker diarization and each step can be accomplished using different models. The steps involved in the speaker diarization include voice activity detection, feature extraction, segmentation, embedding extraction, and clustering. Voice detection can be achieved using Simulink in Matlab, software such as Audacity, Webrtcvad, or other deep learning methods. Further, mel-frequency cepstral coefficients (MFCC) and linear predictive cepstral coefficients (LPCC) are well-known methods available for speech feature extraction. Additionally, segmentation can be achieved using metric-based approaches or by using deep neural networks. There are several frameworks available by Python for the purpose of embedding extraction based on the type of vectors ...