Vocal Performance Evaluation System Based on Fuzzy Inference System

In this era, vocal performance evaluation has become a complex task as it based on subjective judgment and varying criteria. Previous researchers have utilized various traditional methods such as expert scoring and neural networks but still had issues which includes lack of consistency, transparency, interpretability and uncertainties. Therefore, this research proposes a vocal performance evaluation system based on a Fuzzy Inference System (FIS). Firstly, the data is collected from expert evaluations of vocal music performances which contains the evaluation parameters. Then, the collected data is pre-processed with fuzzy logic for handling expert inconsistencies by applying logical rules as well as converts subjective evaluations into standardized fuzzy scores. Next, the obtained data is fed into the Bidirectional Long Short-Term Memory (Bi-LSTM) model for extraction of temporal dependencies and hidden patterns. Finally, the FIS system is utilized to evaluate the final score of vocal performances. The proposed Bi-LSTM-FIS acquired greater results with Mean Squared Error (MSE) of 0.0245 and accuracy of 96.7% when compared to the existing Back Propagation Neural Network (BPNN).