Music Instrument Recognition using Deep Learning
Shubham S. Dubey, Ved V. Hanamshet, Mukesh D. Patil, Dayanand Dhongade · 2023
Music plays a significant role in our daily lives for a vast majority of people worldwide. Various parameters can be used to classify music, including genre, performer, or composer. Automatic musical instrument recognition is crucial for complex tasks such as melody extraction, music information retrieval, and identifying dominant instruments from audio. Recognizing instruments automatically is particularly essential for organizing multiple tracks in an audio clip since manual recognition can be a tedious process. This is where automatic musical instrument recognition comes in handy. It involves detecting and categorizing active instruments, which is the focus of this research. To enhance musical instrument classification, we suggest employing a deep Convolutional Neural Network algorithm designed to efficiently identify musical instruments. Our algorithm includes convolutional layers, and we utilize the Mel spectrogram and Mel Frequency Cepstral Coefficient (MFCC) along with the Convolutional Neural Network to identify instruments present in the Medley Solos DB dataset. This dataset contains over 21,931 audio clips of eight instruments. We perform training based on the model described above, and our testing results show that our suggested method considerably recognizes instruments and achieves better accuracy than the existing approach.