A Hybrid Method for Speech Noise Reduction Using Log-MMSE

Noise reduction is a significant field that appears in many aspects of life. There are many methods for speech noise reduction, especially for stationary noise. This paper talks about a new way to lower speech noise that combines the Log-MMSE (Logarithmic Minimum Mean Square Error) algorithm, which is used to improve speech signals that have been messed up by noise, with an adaptive Wiener filter with a decision-directed (DD) approach. This filter lowers musical noise and predicts the time-varying noise spectrum, which results in a better signal-to-noise ratio. The frame delay issue resulting from DD was resolved by utilizing the Two-Step Noise Reduction (TSNR) technique to reconstruct the harmonic structure of the voice signal that was distorted or missed during the processing and boost speech quality and intelligibility in loud circumstances. Harmonic Regeneration Noise Reduction (HRNR) was used. In this paper, we will investigate these methods in the field of stationary and non-stationary noise. The proposed method was evaluated using different techniques like SNR, perceptual evaluation of speech quality (PESQ), and short-time objective intelligibility (STOI). The proposed approach produced acceptable results in PESQ and STOI, with a considerable increase in the percentage of (SNR), where the percentage of our improvement reached 63.49% in the results of (SNR).