This study proposes a style-aware data augmentation framework that combines rule-based design with statistical constraints, applied to small-scale, highly constrained Jiangnan symbolic music generation tasks. By leveraging YNote representation, fixed rhythmic frameworks, and Markov-style local transition statistics, we systematically expand the training data while maintaining musical structural plausibility. Using the augmented data, we fine-tune a GPT-2 model to analyze how different training data scales affect generation behavior. Experimental results show that Bilingual Evaluation Understudy (BLEU) -based reference-overlap metrics exhibit only minor fluctuations across different training scales and are insufficient to directly reflect style improvement. In contrast, Kullback-Leibler (KL) divergence and bigram statistics effectively characterize the style consistency of the generated set in terms of overall distribution proximity and local transition plausibility. Further analysis indicates that a medium-scale training set (approximately 3,000–12,000 samples) achieves the best balance between distribution consistency and transition coverage, whereas excessive augmentation may lead to distribution calibration drift. Overall, the study demonstrates that data augmentation has a positive but non-monotonic effect on style consistency, highlighting the need for carefully designed augmentation strategies in highly constrained symbolic music generation tasks.
Yung-Ching Tseng, Yu-Chia Wang, Yung-An Chen et al.· International Conference on...· 0 citations
As the application of Unmanned Aerial Vehicles (UAVs) continues to expand globally, the operational health of propulsion components such as brushless motors is critical for ensuring flight safety. Traditional inspection typically relies on manual auditory diagnostics; however, this method is inherently subjective. Because human hearing sensitivity fluctuates across different frequencies, significant discrepancies often exist between objective sound pressure levels and human perception—especially at frequency extremes. Consequently, the reliability and consistency of such auditory-based fault detection are frequently scrutinized. This research establishes an objective, multi-dimensional acoustic feature extraction framework to provide scientific, quantified data that supports human judgment, thereby enhancing diagnostic accuracy. The methodology integrates time-domain, Envelope Analysis, and Time Synchronous Averaging (TSA) techniques to extract key signal features. Analysis of a specific audio sample revealed a signal dominated by intense high-frequency noise peaking at 5,871.09 Hz, exhibiting sharp impulsive characteristics with a Crest Factor (CF) of 4.23. Following TSA processing, asynchronous noise was attenuated by approximately 84.3%, successfully isolating a periodic impact signal at 40.28 Hz, which is precisely synchronous with the shaft rotation speed. The resulting CF of 3.31 confirms the presence of regular, persistent impacts, suggesting potential bearing looseness. The proposed framework effectively isolates weak, fault-related signals from high-intensity noise environments. These objective, quantified results provide a robust scientific basis to assist operators in making more consistent and precise assessments of UAV health. Future research will focus on expanding the experimental dataset and integrating machine learning models to develop a fully automated diagnostic system.
Yan An Lin, Ming-Lang Yeh, Yu-Cheng Lin· AHFE International· 0 citations
