A temporal convolutional network for sequence based performance assessment in artificial intelligence augmented ethnic music education

Ethnic music education is a vital vehicle for cultural heritage, yet traditional teaching relies heavily on teachers’ subjective judgment, making it difficult to objectively quantify learners’ performance development across dimensions such as pitch accuracy, rhythmic accuracy, dynamics control, and expressiveness—a challenge especially pronounced in culturally specific genres such as Hua’er, Jiangnan sizhu, and Mongolian music. To address this, this paper proposes ProgressEthno, a Temporal Convolutional Network (TCN)-based model for sequence-based performance assessment in AI-augmented ethnic music education. Building on a dilated causal convolutional backbone, ProgressEthno incorporates a Style-Aware Feature Fusion (SAFF) module and a Multi-task Progression Scoring Head (MPSH), jointly modeling style-conditioned representations and multidimensional learner-performance patterns. The model is trained and validated on a traditional Chinese folk music dataset (2374 samples, 78-dimensional MFCC features) and a music teaching effectiveness dataset (2000 student records). Experimental results show that ProgressEthno achieves 96.8% accuracy and 96.5% macro-F1 on style classification, and 93.8% accuracy with 93.4% macro-F1 on teaching effectiveness classification, outperforming LSTM, GRU, standard TCN, and Transformer baselines, while maintaining 89.7% teaching-effectiveness classification accuracy under the high-noise condition. Ablation studies confirm the contributions of the SAFF and MPSH modules. These results demonstrate the potential of ProgressEthno as a style-aware, sequence-based framework for intelligent assessment in ethnic music education, while genuine long-term learner progression tracking remains to be validated using prospectively collected longitudinal data.