HYBRID THRESHOLD SPEECH ENHANCEMENT SCHEME USING TEO AND WAVELET COEFFICIENTS Latha R a , Suhas A R a* , B P Pradeep Kumar a , M.Mohammed Ibrahim b and Sathiyapriya V c a Department of Electronics and Communication Engineering, HKBK College of Engineering, Bangalore, India b Department of Computer Science and Engineering, Sona College of Technology, Salem, India c Department of Computer Science and Engineering, Knowledge Institute of Technology, Salem, India Abstract - Speech Enhancement (SE) aims to improve the quality of degraded speech while maintaining its intelligibility. The Wavelet Transform (WT) has become a powerful tool of signal analysis thereby widely used in signal detection and signal denoising. In this paper, we propose an effective means of SE by a hybrid threshold scheme using WT. The proposed methodology looks into both falling the noise and preserving edges of the speech signal unlike the conventional Hybrid Threshold (HT) and Soft Threshold (ST) in the wavelet domain. The threshold value in the wavelet domain is maintained constant for all sub-bands of the signal which reduces denoising efficiency. A novel speech augmentation technique built on the wavelet onsets and time adaption of introduced by calculating wavelet coefficients of the Teager Energy. Performance analysis of speech enhancement techniques using Wavelet coefficients and Teager Energy Operator (TEO) with hybrid threshold method is done. The experiment is carried out for speech data with various values of SNR vacillating from -10 to +10 db with Additive White Gaussian Noise (AWGN). Keywords - Teager Energy Operator (TEO); Additive White Gaussian Noise (AWGN); Wavelet Transform (WT); Speech Enhancement (SE). I. INTRODUCTION Speech samples are typically contaminated by environmental noises, which significantly lowers the signal quality and makes understanding speech signals more challenging. So, it is impossible to exaggerate the value of voice enhancement methods for improving speech and communication applications. A collection of signal processing [1] methods known as speech enhancement work to lessen noise and improve the quality of speech. Voice recognition and speaker identification systems, mobile communication, and hearing aids are just a few areas where speech augmentation is a significant issue. For these applications to produce precise and dependable results, it is crucial to extract clean voice signals from loud situations. Overall, speech enhancement is a significant area of study and development in signal processing since it is essential to enhancing the quality and dependability of speech and communication applications [2]. Speech enhancement approaches aim to increase the quality and clarity of voice signals by removing undesirable noise while retaining the original signal's features. Nevertheless, accomplishing both goals at once may be difficult, and the effectiveness of speech enhancement techniques [3] depends on fulfilling both objective and subjective requirements. The noise type and characteristics are important factors that significantly affect the performance of speech enhancement techniques and limit their applicability. Speech de-noising is a field of engineering that focuses on developing techniques to extract the original speech signal from noisy recordings that various types of noise have corrupted. The noise can take many forms, such as white, red, babble, and other environmental noises. In recent years, removing noise from speech signals has become an area of great interest for researchers in speech processing. Analyzing signals according to their scale is the basic tenet of wavelets. Wavelet may divide a signal into scales that correspond to distinct frequency bands, and at each scale, it is possible to pinpoint the position of the signal's immediate characteristics. For signal de-noising, this feature can be used. [4]. Additionally, wavelets can help reduce the data needed to analyze a signal. Recently, various wavelet-based techniques have been proposed for the cause of speech denoising. The technique is primarily based on the threshold within the sign that each wavelet coefficient of the signal is as compared to a given threshold; if the coefficient is smaller than the edge, then it's far set to 0; otherwise, it's miles saved or slightly reduced in amplitude. Soft and Hard threshold schemes are used for denoising the signals as a compromise for soft and hard threshold schemes which lack keeping edges and de-noising technique effectively hard threshold technique is added. Using Wavelets to dispose of noise from a sign calls for figuring out which additives include the noise, after which reconstructing the signal without the ones additives. Without needing to know the noise level beforehand, a novel method for removing noise from voice signals has been developed that makes 2023 Second IEEE Sponsored International Conference on ICEEICT 2023