Speaker Identification From a Tribal Language Using Several Learning Techniques

A speaker may be recognized by their voice, whether over the phone or by any other digital device. Speaker identification and other audio biometric identification methods have been created to acquire this innate human competence. The article explores machine learning classifiers' efficacy in identifying speakers, a critical component of modern digital security systems. Speaker identification must be precise and dependable since voice-activated interfaces increasingly integrate into technology. This study offers a thorough analysis and comparison of different classifiers. We use the KUI dataset to identify speakers. We obtained Mel-frequency cepstral coefficients (MFCCs) and used them as features for training and evaluating the classifiers. A model's accuracy, precision, recall, and Fl-score were used to assess its performance. According to the findings, XGBoost performed better than any other classifier. These results highlight the efficiency of distance-based and ensemble classifiers in managing intricate patterns for speaker identification. Apart from offering guidance on the appropriate classifiers for speaker identification, the paper establishes the framework for further research concerning the effect of tribal dataset diversity on efficiency in hybrid models. The conclusions substantially contribute to the area by setting the standard for creating sophisticated speaker identification systems.