A Comprehensive Tibetan Folk Song Dataset: Integrating Emotion Classification and Music Information Retrieval in Non-Western Music

This study proposes a systematic method for constructing and analyzing a Tibetan folk song dataset, aiming to provide data support and technical implementation for the classification and sentiment analysis of ethnic music. Tibetan folk songs, as an integral part of Tibetan culture, hold a significant place in the hearts of the Tibetan people, reflecting their emotions, beliefs, and daily life. These songs play a vital role in preserving and transmitting Tibetan cultural heritage, often being performed during ceremonies, festivals, and other important events. The emotional and cultural expressions embedded in Tibetan folk songs offer a unique insight into the identity and values of the Tibetan community.The dataset is constructed based on field surveys, literature, and audio-video resources. Standardizing audio formats, optimizing audio quality, and applying sample balancing techniques ensure the consistency and applicability of the data. Building on this, audio spectral features (such as Mel Frequency Cepstral Coefficients, MFCC, which capture short-term power spectra of sound) and deep features (extracted using YAMNet, a deep neural network for audio signal analysis) are utilized from the dataset. Subsequently, music classification and sentiment analysis are performed using machine learning models and deep learning models. The results demonstrate that combining deep features and spectral features significantly outperforms single-feature approaches, with the XGBoost model achieving a classification accuracy of 91.30% when features are fused. Furthermore, sentiment analysis based on the Russell two-dimensional emotion model reveals that the emotions in Tibetan folk songs predominantly fall into the positive-arousal and positive-calm quadrants, highlighting their unique emotional expressions and cultural characteristics. These findings not only validate the dataset’s effectiveness for music classification tasks but also provide a novel perspective for studying the emotional and structural characteristics of ethnic music.