Optimization of Audio Coding Parameters and Adaptive Denoising Using a Convolutionally Enhanced Transformer Framework

Abstract

With the rapid advancement of digital audio technology, this study proposes an intelligent audio processing framework to address two key challenges: storage optimization and adaptive denoising. By modeling the trade-off between sampling rate, bit depth, and compression algorithm, the system recommends optimal encoding parameters for speech and music to balance file size and audio quality. For denoising, an adaptive algorithm based on time-frequency analysis is introduced, which applies targeted strategies according to identified noise types—Wiener filtering for background noise, median filtering with spectral subtraction for burst noise, and band-stop filtering with spectral smoothing for narrowband interference. Experiments on public datasets using ΔSNR, PESQ, and STOI metrics show that the method improves both noise suppression and audio fidelity, with SNR gains of up to 5.11dB. Subjective listening confirms enhanced clarity, and robustness tests reveal stable performance under moderate noise. Overall, the framework outperforms traditional fixed-parameter methods in both efficiency and quality.