Noise Reduction Strategies For Optimising Speech Signal Clarity
TL;DR
Experimental results indicate that CSM accelerates the training of the models while enhancing the speech quality and show that the method effectively denoises noisy speech audio while being both efficient and accurate.
Abstract
Noise suppression that improves speech intelligibility is critical in teleconferencing, Voice over Internet Protocol (VoIP), hearing aids, and speech recognition systems. This work has attempted an adaptive filtering technique for noise suppression to improve the performance of speech signals. In this regard, two of the most popular algorithms, namely the Recursive Least Squares (RLS) algorithm and Normalised Least Mean Squares (NLMS) algorithm, specifically for existing speech distortion enhancement by room noise, will be analysed. This way, such adaptive filtering techniques can help mitigate those disturbances and enhance speech intelligibility and sound reproduction or processing performance. Phase enhancement to improve the perceptual quality of synthesised speech has recently attracted considerable attention among speech researchers. A few researchers have directly integrated phase estimation modules into speech enhancement architecture with complex-valued time-frequency (T-F) domain representations, such as the complex ratio mask (CRM), for which the phase can be computed from the real and imaginary parts of the complex STFT spectrogram. Unfortunately, spectrogram masking violates these consistency constraints, which can lead to artefacts in the extracted component while also unnecessarily widening the solution space. To address this issue, we propose consistency-based constraints with a novel spectrogram permutation masking technique called Consistency Spectrogram Masking (CSM) to improve the complex spectrogram estimation. Experimental results indicate that CSM accelerates the training of the models while enhancing the speech quality. Our experimental results counter this and show that the method effectively denoises noisy speech audio while being both efficient and accurate.