Low Voice Speech Conversion Analysis Using Novel Convolutional Neural Network Compared with K-Nearest Neighbor with Enhanced Accuracy

This study seeks to improve the process of transforming low voice speech into normal speech, by using the novel convolutional neural network algorithm and comparing its effectiveness with the K-Nearest Neighbor algorithm. Utilizing Machine Learning Algorithms to Transform Low Voice Speech into Normal Speech: A Comparative Study between Convolutional Neural Network and K-Nearest Neighbors, Each Employing 20 Samples per Group. In this study, a pretest power analysis was conducted with a predetermined power score of G as 80% and an interval of confidence as 95%. The results indicate that utilizing the K-Nearest Neighbor and Convolutional Neural Network algorithms for converting low voice speech to normal speech yielded accuracies of 93.89% and 95.89%, respectively. Statistical analysis, through independent sample t-tests, revealed to a considerable dissimilarity in accuracy between both algorithms (p < 0.05), with a boundary of 0.002. The findings suggest that the Convolutional Neural Network approach outperforms the K-Nearest Neighbor technique in this context.