Real-Time Mobile Music Note and Instrument Recognition: A Unified Deep Learning vs. Classical ML Benchmark on MusicNet and NSynth

The rapid growth of artificial intelligence and mobile computing has enabled real-time music analysis; however, accurate musical note and instrument recognition on mobile devices remains challenging due to limited computational resources, noisy audio, and strict latency constraints. This paper presents Instrumaster, a unified mobile framework for real-time musical note and instrument recognition that integrates robust audio preprocessing, feature engineering, and efficient inference. Musical note recognition is evaluated using LSTM, CNN, Feedforward Neural Network (FNN), and Logistic Regression models, while instrument recognition is performed using a Multi-Layer Perceptron (MLP). Experiments conducted on the MusicNet and NSynth datasets demonstrate that sequential models effectively capture temporal dependencies, while classical machine learning approaches can achieve competitive performance with significantly lower computational complexity. Notably, Logistic Regression achieves strong accuracy under limited data conditions, highlighting the importance of informed model selection for mobile deployment. Overall, the results provide practical insights into accuracy–efficiency trade-offs and establish a reference framework for designing reliable and real-time mobile music recognition systems.