Speech Enhancement Techniques and its Implementation

Speech enhancement aims to improve quality and intelligibility. Quality refers to the amount of noise free in speech and intelligibility refers to the percentage number of words understand in the sentence. Speech enhancement involves noise estimation as crucial part. Noise estimation approaches and their implementation using MATLAB is discussed in this work. Basic noise estimation includes estimation of noise from first frames and then speech presence and absence based approaches are discussed. Also the statistical based approaches for speech enhancement plays important role in speech enhancement. In this work traditional statistical estimation methods and their implementation using MATLAB is discussed. In this work different databases (clean, noisy speech samples) available for speech enhancement are discussed. To evaluate the speech enhancement algorithms the objective and subjective performance measures available in literature are discussed and the sources for MATLAB implementation is discussed. The performance of speech enhancement methods is compared with help of Signal to Noise Ratio (SNR), Segmental SNR (Seg SNR), Perceptual Evaluation of Speech Quality (PESQ). Also the subjective performance is analyzed using spectrograms.