Ethnic Music Style Migration Algorithm Based on Artificial Intelligence

This paper introduces a novel approach that is designed for the purpose of enhancing the migration of Ethnic Music (EM) styles into diverse musical content using Artificial Intelligence (AI) techniques. EM is considered to be rich in cultural significance. It is characterized by unique scales, rhythms, and instruments, so it presents specific challenges in style transfer, mainly in terms of preserving the authenticity and integrity of traditional sounds while, at the same time, trying to introduce new stylistic elements. The proposed Migration Algorithm (MA) employs a combination of techniques such as Short-Time Fourier Transform (STFT) for preprocessing, Feature Extraction (FE) using Mel scale filtering and Mel-Frequency Cepstral Coefficients (MFCC), and feature transformation modified convolutional neural networks (STLCNet for style and CONCNet for content). The output from both networks is combined using the mathematical theory of evidence, and its output spectrogram is reconstructed into an audio signal using the phase gradient heap integration (PGHI) algorithm. An additional 60,000 recordings have been included in the Smithsonian Folkways Collection for testing. The approach succeeded significantly on key metrics, including KL Divergence, Signal-to-Distortion Ratio (SDR), and Signal-to-Noise Ratio (SNR), and the empirical findings proved that type of transfer accuracy, quality of sound, and text integrity were all properly addressed.