Jagadish S.Jakati et al., International Journal of Emerging Trends in Engineering Research, 8(6), June 2020, 2472 - 2480 2472 ABSTRACT Noise is introduced in speech signal due to various unavoidable reasons which makes the speech less intelligible. De-noising is an ongoing research area from past many decades. De-noising is achieved by different type of filtering techniques. Performance of the particular de-noising algorithm depends on the characteristics of the filter used. In this paper, an efficient speech denoising technique using Discrete Wavelet Transform (DWT) and thresholding is proposed. To get optimal de-noising, multilevel DWT is used. This method separates the noise components present in the noisy speech and further noise components are suppressed by thresholding of DWT coefficients. Clean speech signal is reconstructed by performing Inverse Discrete Wavelet Transform (IDWT). The results show that good quality de-noised signal is obtained using Haar wavelet with Mini-Maxi thresholding technique. On the other hand, low error rate is obtained using dB13 wavelet with Rigsure thresholding technique. DB13 and Sym13 wavelets with Rigsure thresholding technique provide good tradeoff between error rate and the quality of the de-noised signal. Key words: Speech Processing, Discrete Wavelet Transform, Thresholding, Signal to Noise Ratio etc. 1. INTRODUCTION. The processing of speech signal is a part of Digital Signal Processing where digital algorithms are used to process the speech. But any speech signal is analogous in nature which is digitized by suitable ADC architecture [1] for before processing. Due to the various design issues, noises are introduced in the digitized speech signal. Normally the noises can be added in the signal due to some external noise sources and sometime due to the effect of communication channel. All those above mentioned noise degrades the speech quality depends upon the intensity of noise. As a result it is essential to develop algorithmic model to minimize the noises from noisy speech signal. For this denoising normally the use of filter banks are popular among most of the existing filter bank, the filter band of Discrete Wavelet Transform (DWT) [2] has more capabilities to denoise any signal. In this paper, an efficient speech denoising algorithm is prepared which denoise any speech signal using multilevel DWT and soft thresholding operation. 2. LITERATURE SURVEYS The existing techniques used to denoise speech signal are explained in this section briefly. In this paper [3], a new method has been proposed that deals with dual channel speech enhancement methods utilize the coherence function determine from the input signals without prior noise statistics. Discrete Wavelet Transform (DWT) is make use of a gain function. This method gave the good result when speech is perverted by various noise types when applied in an domain where interfering speakers are present. In this paper [4], they presented the usage of denoising wavelet on speech input of MFCC (Mel Frequency Cepstral Coefficient) feature extraction method. The denoising process using wavelet transformation is used to enhance the MFCC shows on noisy signals. They used 120 speech data, with 30 data were used as the reference, and the other 90 were used as the testing data. These methods using wavelet transformation are able to boost the accuracy of the speech recognition system on input signals with SNR of 0-10 dB. In this paper [5], to remove the noise on the audio signal Discrete wavelet transform ¸ based algorithm has been used. Both hard and soft thresholding are used for denoising. This method gave the good and efficient results and can be used real-time processing. In this paper [6], to estimate the nature of noise power spectrum of the E-DATE algorithm is subsequent by using DWT instead of STFT. The novel method recovered STFT of the input speech signal by Efficient Speech De-noising Algorithm using Multi-level Discrete Wavelet Transform and Thresholding Jagadish S.Jakati 1 , Shridhar S.Kuntoji 2 . 1 Assistant Professor, Department of Electronics & Communication Engg, S. G. Balekundri Institute of Technology Belagavi, VTU Research Scholar Karanataka (State), India, jagadishjs30@gmail.com 2 Professor, Department of Electronics & Communication Engg, Basaveshwar Engineering College Bagalkot, VTU Research Supervisor, Karanataka (State), India, shridhar.ece@gmail.com ISSN 2347 - 3983 Volume 8. No. 6, June 2020 International Journal of Emerging Trends in Engineering Research Available Online at http://www.warse.org/IJETER/static/pdf/file/ijeter43862020.pdf https://doi.org/10.30534/ijeter/2020/43862020