<![CDATA[The exponential growth of music libraries on streaming services has led to information overload, making it difficult for listeners to find works that match their preferences. The main issue faced is the cold-start problem, which is the failure of the system to recommend new tracks or specific genres such as Psychedelic Rock due to a lack of historical interaction data. This study aims to build a content-based filtering recommendation system architecture to overcome this problem. The technical procedure begins with the conversion of analog audio signals into digital spectral representations using the Librosa module. Feature extraction focuses on the parameters of Mel-Frequency Cepstral Coefficients (MFCC), Spectral Centroid, and Zero Crossing Rate (ZCR). Next, the similarity level between music entities is calculated using the Cosine Similarity metric. Testing on a data corpus consisting of 100 song samples shows satisfactory system performance with 88% accuracy and 86% precision. These results validate that the combination of audio feature extraction and the cosine similarity algorithm is effective in providing accurate recommendations without relying on user history.]]>
Copyrights © 2026
