RESPIRATORY LUNG SOUNDS RECOGNITION USING NEURAL NETWORKS

This work offers a sophisticated method for identifying lung auscultation sounds that makes use of neural networks, chroma characteristics, and Mel-frequency Cepstral Coefficients (MFCC). A crucial diagnostic technique for determining respiratory disorders, lung auscultation frequently depends on the skill of medical practitioners to decipher minute sound patterns. However, early identification and therapy can be significantly aided by automated methods that correctly categorize these noises. In order to accomplish this, we used MFCC, which focuses on the essential frequency ranges for lung sounds and efficiently mimics how humans perceive auditory signals by capturing the power spectrum of sounds. Additionally, harmonic characteristics that might be suggestive of particular lung disorders were captured using chroma features, which characterize the tonal content of audio signals. A neural network that was created to categorize lung sounds into different diagnostic groups, including normal breathing, wheezing, crackles, and other aberrant respiratory sounds, was then fed these properties. The neural network achieved good classification accuracy by learning intricate patterns and correlations within the MFCC and Chroma features after being trained on an extensive dataset of lung sounds. The accuracy of lung sound diagnosis can be increased using this automated method, which may result in earlier identification of respiratory disorders and better patient outcomes