http://javad.dev/tags/audio/Recent content in Audio on Javad Rajabzadeh | Senior Software EngineerHugoen-usSun, 19 Jul 2026 22:30:00 +0330http://javad.dev/posts/omlsa-speech-enhancement-postfiltering/Sun, 19 Jul 2026 22:30:00 +0330http://javad.dev/posts/omlsa-speech-enhancement-postfiltering/<p>In digital signal processing and speech enhancement, spatial filtering techniques (such as Multi-channel Wiener Filtering or beamforming) effectively attenuate directional noise sources. However, residual non-stationary noise, diffuse background acoustic clutter, and reverberation often linger in the processed audio.</p> <p>Applying traditional spectral subtraction or hard noise gate suppression to clean this residual noise causes severe <strong>musical noise artifacts</strong>—random, isolated spectral peaks that destroy human speech intelligibility and sound unnaturally harsh.</p> <p>In this post, I explore the mathematical formulation and practical implementation of <strong>Optimally-Modified Log-Spectral Amplitude (OM-LSA)</strong> speech enhancement—a single-channel post-filtering technique designed by Israel Cohen and Baruch Berdugo that minimizes log-spectral amplitude estimation errors while preserving natural speech quality.</p>