Optimizing Audio Compression Through Entropy-Controlled Dithering

This research explores distortion-controlled dithering in audio compression, focusing on TPDF dithering and its modifications. By adding noise to a signal before quantization, dithering reduces perceptual artifacts like graininess and ringing. Standard and modified TPDF dithering, with and without noise shaping, were evaluated using perceptual metrics (VISQOL and STOI) and entropy on audio samples of varying complexity. Results show that noise shaping improves simple audio quality but is less effective for complex signals, while modified TPDF without noise shaping underperformed. Optimization of the alpha parameter, which controls dither amplitude, revealed consistent trends across similar signal types. The findings highlight TPDF dithering’s adaptability in achieving near-lossless audio quality with efficient storage and supporting applications in machine learning, telecommunications, and assistive devices. A practical implementation as a Digital Audio Workstation plugin introduces customizable dithering controls, laying a foundation for advancements in audio compression algorithms.