Certain investigations on speech enhancement algorithms for efficient speech recognition

Abstract

Recent developments in speech processing have enabled Artificial Intelligence to work on systems to extract, analyse and understand valuable information from audio data. Speech processing is extracting the features from speech, eliminating the noise present in the speech and improvising the speech quality. Intelligent speech processing has increased in attraction across all voice-controlled applications and academics. There is a wide range of applications using speech processing, including smart home, health-care, retail, entertainment and automobiles. Many applications of speech processing mainly focus on speech quality improvement. Speech enhancement and recognition are essential and significant tasks in speech processing. Though many advancements exist in speech processing, efficient identification of external noise present in the speech for accurate elimination of noise from the corrupted speech remains challenging. newlineThe speech enhancement model can be developed using traditional, hybrid and optimised speech enhancement approaches. The speech enhancement process involves two main steps: noise segment identification and enhancing speech quality by eliminating the extracted noise from the input speech. The noise segment can be extracted accurately from the original speech signal during feature extraction. The traditional filter-based approaches for speech processing struggle with several challenges like the type of noise, the accent of the speaker and the voice of the speakers. Hence the accuracy in speech enhancement was low and suffered from increased word error rates. newline newline