International Journal of Science, Engineering and TechnologyAuthors: Ishwari Gadewar, Mitali Umbarje, Vedika Thakur, Prof. S. R. Warhade
Abstract: Accurate identification of musical key tones in har-monium performance is essential for effective learning, tuning, and real-time feedback in Indian classical music. However, existing pitch detection tools are primarily designed for Western musical scales and fail to capture the microtonal nuances and timbre characteristics of the harmonium. Additionally, manual tone identification is time-consuming and prone to human error, especially for novice learners. This paper proposes a novel Auto-matic Key Tone Detection System for Harmonium using Artificial Intelligence (AI) and Machine Learning (ML) techniques. The system captures live audio input and processes it through a real-time pipeline involving signal preprocessing, feature extraction using Mel-Frequency Cepstral Coefficients (MFCCs) and log-Mel spectrograms, and classification using a lightweight Convo-lutional Neural Network (CNN) model. A key design aspect is the system’s low-latency architecture, enabling real-time detection with high accuracy while maintaining computational efficiency suitable for standard desktop environments. The proposed model is trained on a curated dataset of harmonium tones across multiple shrutis and octaves, augmented to improve robustness under varying acoustic conditions. The system provides real-time output displaying the detected swara, octave, and confidence score through a user-friendly interface. By bridging the gap between traditional musical practice and modern AI-driven tools, this solution aims to enhance learning efficiency, improve tuning accuracy, and support musicians in real-time performance environments.
DOI: http://doi.org/10.5281/zenodo.20957304
