Audio Recognition System using Fourier Transform

This paper presents an audio recognition system aimed at improving robustness in noisy environments. Our approach employs a signature-based recognition methodology that utilizes minimal pre-processing through a Gaussian convolution alongside Fast Fourier Transform (FFT) to extract distinctive audio signatures. The system is designed to optimize performance in low Signal-to-Noise Ratio (SNR) conditions, demonstrating significant accuracy improvements compared to existing models. Notably, our system achieves an accuracy of 85.7% at a 50% SNR, substantially outperforming traditional methods that exhibit only 57% accuracy under similar conditions. Furthermore, the architecture exhibits linear scalability in both build initiation time and memory requirements relative to the size of audio samples, while matching times remain constant with respect to the database size.