MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202641101828 A) filed by St. Josephs College Of Engineering; and Ms. S P Vedavalli on August 22, 2026, for A Cnn-Based Music Instrument Recognition System For Automated Identification Of Musical Instruments From Audio Recordings Using Deep Learning And Audio Signal Processing.

Inventors include St. Josephs College Of Engineering; Ms. S P Vedavalli; Anish T; and Antin Shanio A.

The application for the patent was published on August 28, 2026, under issue no. 35/2026.

Abstract: This project presents an intelligent CNN-Based Music Instrument Recognition System designed to automate the identification of musical instruments from audio recordings. Traditional manual tagging is often inaccurate and time-consuming, while conventional signal processing methods frequently lack the precision required for complex audio analysis. The system utilizes a structured machine learning pipeline where raw audio signals (WAV/MP3) are preprocessed using the Librosa library and transformed into Mel- Frequency Cepstral Coefficients (MFCC). These features provide a compact representation of frequency patterns aligned with human auditory perception. A Convolutional Neural Network (CNN), trained on the IRMAS dataset, is employed to classify these features into specific instrument families. The deep learning model architecture includes multiple convolutional layers for feature extraction, dropout layers to prevent overfitting, and a sigmoid activation layer to support multi-label recognition. Integrated into a user-friendly Streamlit web application, the system provides interactive visualizations, including waveforms and spectrograms, to offer interpretable feedback on prediction results. Experimental results demonstrate high training accuracy (reaching approximately 98.6%), though validation accuracy (approximately 54.7%) suggests ongoing challenges with model generalization in diverse audio environments. This research highlights the efficacy of combining deep learning with advanced audio signal processing for digital music analysis and education.

Disclaimer: Curated by HT Syndication.