Research on Intelligent Coding Optimization and Speech Denoising & Enhancement Methods for Audio Processing

Abstract

To address the trade-off between audio quality and file size as well as the demand for enhancing noisy speech intelligibility, this paper proposes a full-process audio processing scheme integrating intelligent analysis, adaptive coding optimization, speech denoising and quantitative evaluation. First, 34 music and speech samples with diverse coding parameters are used to construct the Quality-Size Index (QSI), and combined with a random forest regression model, bit rate and compression algorithm are identified as the core factors affecting coding performance. Second, a random forest classifier based on audio time-frequency features achieves 100% accuracy in music-speech classification, enabling adaptive coding optimization via matching optimal parameters for different audio types. Finally, an optimized spectral subtraction-Wiener filtering cascade denoising algorithm is designed for non-reference noisy speech. Experimental results show that the adaptive coding scheme effectively balances audio quality and file size; the proposed denoising algorithm improves the SNR of two speech samples by 6.22 dB and 5.51 dB respectively, significantly reduces spectral flatness, eliminates musical noise and enhances speech intelligibility. This scheme provides an intelligent technical solution for audio coding and speech denoising with high practical application value.