Optimization-Driven Deep Learning Framework for Ethnic Instrumental Music Style Recognition and Cross-Cultural Semantic Dissemination

Abstract

To enhance the recognition accuracy and dissemination adaptability of ethnic musical instrument styles in multiple contexts, this paper proposes an optimization algorithm-driven deep learning system framework for the recognition of ethnic musical instrument styles and cross-cultural semantic dissemination. The research first constructs a database containing multi-ethnic instrumental audio and three-layer cultural semantic labels, and uses CNN, LSTM and Transformer to build a multi-channel fusion model to achieve collaborative modeling of timbre, rhythm and structural information. To optimize the model structure and parameter configuration, Particle swarm Optimization (PSO) is introduced for network structure search, and Bayesian optimization is combined to fine-tune key hyperparameters such as Dropout rate and learning rate. The system was trained and deployed on the NVIDIA A100 cluster, and a 50% cross-validation was conducted using Top-1 Accuracy, Macro F1-score, and Top-3 Accuracy as evaluation metrics. The results show that the optimization strategy improves the Top-1 Accuracy by 6.2% compared with the baseline model, and the Top-3 Accuracy reaches 91.4%. The system further integrates the style semantic mapping mechanism with the human-computer interaction recommendation interface, achieving style content retrieval and dissemination path guidance based on users' emotions and cultural cognitive preferences, significantly enhancing the system's cultural adaptability and user comprehension. The research integrates artificial intelligence with music information processing technology, providing a scalable system solution for the intelligent recognition and global dissemination of ethnic Musical Instruments.