Publication

Abstract : This paper introduces a hybrid music recommendation framework that integrates advanced speech processing techniques—specifically Mel-Frequency Cepstral Coefficients (MFCCs) and Dynamic Time Warping (DTW)—to extract and compare rich acoustic representations of music. In contrast to traditional systems that rely heavily on user behavior or metadata, our approach treats music as a perceptually complex audio signal. MFCCs are employed to capture spectral timbre, while DTW ensures time-aligned similarity matching, accommodating tempo and structural variations. These methods, originally rooted in speech recognition, enable fine-grained analysis of musical patterns beyond surface-level tags. To complement this contentaware layer, we incorporate a genre-based recommendation component that simulates collaborative filtering via supervised classification. The resulting hybrid system dynamically balances acoustic similarity with inferred genre preferences, improving recommendation relevance—especially in cold-start scenarios with limited user history. Our framework demonstrates that combining low-level acoustic feature analysis with adaptive learning not only mitigates popularity bias but also enables scalable, high-resolution music recommendation grounded in the intrinsic properties of the audio signal.