Research on Automatic Identification and Digital Preservation of Musical Intangible Cultural Heritage Performance Techniques Based on Deep Learning

The preservation of musical intangible cultural heritage requires effective approaches for identifying and documenting complex performance techniques that are often transmitted through tacit knowledge and oral instruction. To overcome the limitations of manual annotation, low-dimensional recording methods, and insufficient quantitative analysis, this study proposes a deep-learning-based automatic recognition framework for musical intangible cultural heritage performance techniques. A multi-source dataset containing representative performance recordings from guzheng, Nanyin, Finnish fiddle, and other traditional musical forms is first established. A hybrid architecture integrating cascaded convolutional gated neural networks and fully convolutional networks is then developed to capture local acoustic features and long-range temporal dependencies. Furthermore, note-onset information is incorporated to improve technique-boundary localization accuracy. Experimental results demonstrate robust recognition performance across diverse musical styles and performance conditions. The proposed framework supports digital archiving, intelligent retrieval, and interactive learning applications while providing methodological references for audio signal processing, pattern recognition, and intelligent cultural-information preservation systems.