The state of the art on ASR systems and feature extraction technique

Abstract

The foremost purpose of a speech enhancement system is to bring out intelligent speech among the mixture of noise, background audio, and the speaker. This can be utilized in intelligent speech transmission, hearing aids, speaker identification, video conferencing, etc. Over the years, many speech enhancement methods have been implemented to process the noisy background and enhance the quality of the speech signal. With the introduction of machine learning, deep learning, and neural networks, this enhancement of speech has become more fruitful. However, the traditional methods like spectral subtraction, wiener filtering, etc., faced the musical noise issue, especially while working on a low SNR signal. While many DNN-based methods are being introduced, denoising and dereverberation are not collectively considered that often. This article provides an overview of the research on speech enhancement techniques in the last several years. The various challenges and future research ideas are also covered, considering the present methods of speech enhancement.