Transcription
S SANGEETHA AND P KANNAN: DESIGN AND ANALYSIS OF DIGITAL FILTERS FOR SPEECH SIGNALS USING MULTIRATE SIGNAL PROCESSINGDOI: 10.21917/ijme.2018.0086DESIGN AND ANALYSIS OF DIGITAL FILTERS FOR SPEECH SIGNALS USINGMULTIRATE SIGNAL PROCESSINGS. Sangeetha and P. KannanDepartment of Electronics and Communication Engineering, PET Engineering College, IndiaHanning and Blackman window. For investigation, first of all asignal out of audio frequency range is chosen and after that a littlepartition of this signal is extracted utilizing framing technique.The resulting signal frame is gone through Hamming, Hanningand Blackman window and their particular power spectraldensities are calculated. To investigate power content of signalFFT is utilized. It can be obtained from the simulated results thatthe Blackman window contains almost double power whencontrasted with Hamming and Hanning window. So for longdistance communication Blackman window is utilized. Thedrawback of this paper is that the losses happen while travellinglong distance.Babu, et al. [14] utilized windowing techniques and theperformance of Hamming, Hanning and Blackman windows aremainly compared depending on their magnitude response, phaseresponse for designing the FIR low pass and high pass filters usingmatlab. The responses of FIR low pass, high pass filters can beobtained from its design. In this paper, on looking at thesimulation results utilizing different windows, it is observed thatthe Blackman window creates better results among them and theresponse of Blackman window are more smooth and perfect.Singh [5] built up speech signal analysis strategy taking intoaccount Fast Fourier Transform (FFT) and Linear PredictiveCoding (LPC). In this paper five samples of single word are takenby same individual. These samples are examined utilizing FFTand LPC as a part of matlab. After investigation variousparameters of tests are gotten for FFT and LPC range exclusively.The primary restriction of this paper is that the spectrum analysisis a complex process of decomposing the speech signal intosimilar parts.Naik, et al. [6] exhibited a very low bit rate speech codertaking into account sub-band coding technique. The audiblefrequency range 20Hz-20kHz is split into frequency sub-bandsutilizing a bank of finite impulse response filter. Then the outputof every filter is sampled and encoded. At the output, the signalsare de-multiplexed, decoded and demodulated and afterwardssummed to recreate the signal. This paper mainly focusing thecomparison of correlation values for various clean speech signalsand correlation values for after adding high amplitude noise to thesame speech signals. Taking correlation tests demonstrate that itsexecution is fulfilling.Podder, et al. [7] proposed windowing techniques for thecomparison of performance of Hamming, Hanning and Blackmanwindow based on their magnitude response, phase response andequivalent noise bandwidth in time and frequency domainutilizing matlab simulation. To observe the responses, FIR filterof low pass, high pass, band pass and band stop filters have beendesigned. Looking simulation consequences of various windows,Blackman window has best execution among them and theresponse of the Blackman window is more smooth and perfectAbstractDigital filters provide an important role in the world of communication.This paper proposes the design of digital filters for audio applicationusing multi rate signal processing. One of the important applicationsin multi rate signal processing is sub band coding. The main objectiveof this paper is to analyze various techniques for designing digitalfilters for speech signals. Additive White Gaussian Noise is added withthe input speech signal. The input speech signal spectrum is dividedinto frequency sub-bands using down sampling by a factor 2. Varioustransforms like FFT, FWHT and DWT are applied to the signal and itssub bands. Then the low pass and high pass FIR filters are designedand implemented using windowing techniques and IIR filters aredesigned and implemented using Butterworth and Chebyshev filters.Finally quantization is performed on the filter coefficients of signal andits sub bands. The performances of digital filters are measured bycalculating Signal to Quantization Noise Ratio. From the performancemeasures this paper concludes that, which filtering technique is mostsuitable for designing digital filters for speech signals.Keywords:Digital Filters, Sub Band Coding, FIR, IIR, DWT, FFT, FWHT,Quantization, SQNR1. INTRODUCTIONSpeech is the most basic and preferred means ofcommunication among humans. In speech processing, a filterremoves the unwanted signal and allows the desired signal. Filtersmay be analog or digital. Digital filtering is one of the importanttools for digital signal processing applications. Digital filters arecapable of performing that specifications which are extremelydifficult, to achieve with an analog implementation. Multiplefiltering is possible and it can be operated over wide range offrequencies, because the characteristics of digital filters can beeasily changed under software control. Digital filters are classifiedeither as Finite duration impulse response (FIR) filters or Infiniteduration impulse response (IIR) filters, depending on the form ofimpulse response of the system.In this paper, these FIR and IIR filters for speech signals aredesigned and implemented using multi rate signal processing.The systems that employ multiple sampling rates in the processingof signals are called multi rate signal processing systems.Multirate digital signal processing systems use down sampler andup sampler, the two basic sampling rate alteration devices inaddition to conventional elements like adder, multiplier and delayto change the sampling rate of a digital signal.2. LITERATURE SURVEYSaini, et al. [4] displayed comparative analysis of speechsignal utilizing different windowing methods such as Hamming,480
ISSN: 2395-1680 (ONLINE)ICTACT JOURNAL ON MICROELECTRONICS, JANUARY 2018, VOLUME: 03, ISSUE: 04when compared with Hamming and Hanning windows. Thefundamental downside is that the Blackman window has higherequivalent noise bandwidth.compress the signal without losing any important information.Some of the techniques are FFT, DWT and FWHT.3. PROPOSED METHODFFT algorithm computes the DFT of a sequence or its inverse.It is an efficient algorithm to compute DFT and its inverse whichcompute the same result quickly. Because DFT requires N2operation for N-point sequence while FFT requires only NlogNoperation.Formula to compute FFT is,3.2.1 Fast Fourier Transform (FFT):The speech signal is taken as the input signal. AWGN is addedwith the input speech signal. The noisy speech signal spectrum isdown sampled into multiple sampling rates using sampling rateconversion. Various transforms like Fast Fourier Transform(FFT), Fast Walsh Hadamard Transform (FWHT) and DiscreteWavelet Transform (DWT) are applied to the noisy speech signaland its sub bands. The FIR filters are designed and implementedusing different window functions such as Rectangular, Hanning,Hamming, Blackman and Kaiser windows and the IIR filters aredesigned using Butterworth and Chebyshev filters. ThenQuantization is applied to the filter coefficients. Finally theperformance of the filter coefficients are measured based on theSignal to Quantization Ratio (SQNR). The Fig.1 shows the overallprocess of the proposed method.N 1X k xn e i 2 nk / N ; k 0,1,. N 1where,k - index of frequencies, n - index of signal and N – bandwidth.3.2.2 Fast Walsh Hadamard Transform (FWHT):FWHT is an efficient algorithm to compute the WHT (WalshHadamard Transform). FWHT is a divide and conquer algorithmthat recursively breaks down a WHT of size n into smaller WHTof size n/2. WHT is a non-sinusoidal orthogonal transform thatuses Walsh function which is determined from the number of zerocrossing per unit time interval. Walsh function is rectangular orsquare value and real value. It requires only NlogN additions orsubtractions.Formula to compute ormFFTIIRSamplingDWTFiltersPerformancemeasuresN 1n 1m 0i 0X [k ] x[m] ( 1) ( ki ki 1 ) mn 1 iFIRMultiRate(1)n 0FWHT(2)3.2.3 Discrete Wavelet Transform (DWT):In numerical analysis and functional analysis, a DiscreteWavelet Transform is any wavelet transform for which thewavelets are discretely sampled. A key advantage of wavelettransform over Fourier transform is temporal resolution.Temporal resolution-it captures both frequency and locationinformation (location in time).WithquantizationFig.1. Block diagram of proposed method3.1 SAMPLING RATE CONVERSIONThe noisy speech signal spectrum is divided into fourfrequency sub bands by down sampling by a factor of 2. Thebelow Fig.2 shows the down sampler.3.3 FILTERING TECHNIQUESFiltering techniques are useful for separating, extracting andrestoring the signal. It helps to remove the sources of noise. Thereare two types of digital filtering techniques. FIR filtering technique IIR filtering techniqueMFig.2. Downsampler3.3.1 FIR Filter Design:Input Sampling Frequency F 1/TOutput Sampling Frequency F′ F/M 1/MTIn FIR Filter the present output depends on present input andpast input. The impulse response of continuous time signal issampled into impulse response of discrete time signal. There is nofeedback required for implementing FIR filter. Therefore FIRfilters are called Non Recursive Filters. FIR filters are alwaysstable and finite. Here, FIR Filters are designed using WindowingMethods. A Window function is a mathematical function that iszero valued outside of some chosen interval. The basic principleof window function is to calculate hd (n) by the Fourier transformbased on the filter frequency response.3.2 TRANSFORMIn speech processing, the speech signal should be firsttransformed and compressed for further processing. They are notnecessary, but they do make some calculations much simpler andmore convenient. It is possible to do all computation and analysisof a signal in either time domain or the frequency domain.However some operations are much simpler and more intuitive inone than the other. Transforms are the tools to convert timedomain into frequency domain. There are many techniquesavailable which are used to extract the important features andhd (n) 48112 H d(e jw )(3)
S SANGEETHA AND P KANNAN: DESIGN AND ANALYSIS OF DIGITAL FILTERS FOR SPEECH SIGNALS USING MULTIRATE SIGNAL PROCESSING Then h(n) can be obtained by multiplying the windowsequence, w(n) with the hd (n).h ( n ) w( n ) hd ( n )H z h n z n(4)The commonly used filters are, Butterworth filters - No ripples at all, Chebyshev filters - Ripples in pass band or stop band, and Elliptic filter - Ripples in both pass band and stop band.The magnitude function of the Butterworth filter is given by,For low pass filter, hd (n) is given by,sin wc ( n )n For high pass filter, hd (n) is given by,hd ( n ) (5) sin wc ( n )(6)n The common types of windowing methods are, Hamming Window: Hamming Window is the raised cosinewindow which minimizes the maximum side lobe andprovide a height of about one fifth of Hanning Window.hd ( n ) 2n 0.54 0.46cos 0 n N 1w(n) N 1 0otherwise H ( j ) 0 n N 1 2 N 1 c 0.512H ( j ) (7)1 CN2(13)2 P (14)where,N - Order of the filterΩc - Cut off frequency.Ωp - Pass band frequencyCN - cos(Ncos-1x), x 1 (Pass band)CN - cos(Ncos-1x), x 1 (Stop band)(8)3.4 QUANTIZATIONotherwise Blackman Window: Blackman Window is similar toHamming and Hanning Window and the width of the mainlobe is low when compared to Hamming and HanningWindow.w( n ) Quantization is the process of converting Discrete TimeContinuous Amplitude into Discrete Time Discrete Amplitudesignal. There are two types of quantization namely, truncation androundingThe quantization error is given by,e(n) xq(n)–x(n)(15)where,xq(n) - Quantized sample of the signal.x(n) - Unquantized sample of the signal. 2n 4n 0.42 0.5cos 0.08cos N 1 0 n N 1 (9)N 1 0otherwise Rectangular Window: Rectangular Window is the simplestwindow, taking a chunk of the signal without anymodification.3.5 PERFORMANCE MEASURES0 n N 1 1(10)w(n) otherwise 0 Kaiser Window: Kaiser Window maximizes the ratio ofmain lobe energy to side lobe energy. In Kaiser Window fora particular length, the particular parameter controls the sidelobe height.2 2n 1 1 0 n N 1w(n ) N 1 otherwise 01The magnitude square function of the Chebyshev filter isgiven by, Hanning Window: Hanning window is the raised cosinewindow which has minimum stop band attenuation. It can beused to reduce the side lobe while preserving the goodfrequency resolution. Hanning window touches zero at bothends and removes any discontinuities. 2n 0.5 0.5cos w(n ) N 1 0 (12)n 03.5.1 Signal to Quantization Noise Ratio (SQNR):SQNR can be applied to the Quantized signal. SQNR can becalculated between the Normalized Signal power and theQuantization Noise power.Signal to Quantization Noise Ratio (SQNR) is given by, Normalized Signal Power SQNR 10log10 Quantization Noise Power (11)(16)4. RESULTS AND DISCUSSIONSThe experimental results of the proposed methods are shownin the below figures. The voice sample of the male is taken foranalysis of the speech signal from the TIMIT (Texas Instrumentsand Massachusetts Institute of Technology) data base. TIMIT data IIR Filter Design: A filter whose impulse response isinfinite can be considered as an IIR filter and it has afeedback loop. The disadvantages of IIR filter are that theyusually have nonlinear phase. The transfer function of theIIR Filter is given by,482
ISSN: 2395-1680 (ONLINE)ICTACT JOURNAL ON MICROELECTRONICS, JANUARY 2018, VOLUME: 03, ISSUE: 04base was compiled using 630 speakers and each speaker spoke tenutterances giving a total of 6300 sentences.1 has the frequency of 4000Hz, Band 2 has the frequency of2000Hz and so on.4.1 INPUT SIGNAL4.3 TRANSFORMFig.3. Speech signalThe Fig.3 shows the original signal representation of speechsignal which was spoken by a man. The voice is recorded and itis stored as a wave file for further usage in matlab. The frequencyof the input signal is 8000Hz.Fig.6. Various Transforms applied Original signalThe Fig.6 shows the responses of original signal transformedby various transforms techniques such as FFT, FWHT and DWT.4.4 FIR FILTERINGFig.4. Noisy Speech signalThe Fig.4 shows the noisy input signal representation whichis treated as the original signal. Noisy speech signal is the AWGNadded speech signal.4.2 SAMPLING RATE CONVERSIONFig.7. FFT applied Original signal filtered by FIR-LPFFig.5. Sub bands of original signalThe Fig.5 shows the sub bands of original signal which is adown sampled signal of original speech signal by a factor 2. BandFig.8. FFT applied Original signal filtered by FIR-HPF483
S SANGEETHA AND P KANNAN: DESIGN AND ANALYSIS OF DIGITAL FILTERS FOR SPEECH SIGNALS USING MULTIRATE SIGNAL PROCESSINGThe Fig.7 and Fig.8 shows the responses of FIR LPF and HPFdesigned by different windowing such as Hamming, Hanning,Blackman, Rectangular and Kaiser Window techniques of FFTapplied original signal. From these windows Blackman HPFprovides high SQNR than other windowing techniques.Fig.11. DWT applied Original signal filtered by FIR-LPFFig.9. FWHT applied Original signal filtered by FIR-LPFFig.12. DWT applied Original signal filtered by FIR-HPFThe Fig.11 and Fig.12 show the responses of FIR LPF andHPF designed by different windowing such as Hamming,Hanning, Blackman, Rectangular and Kaiser Window techniquesof DWHT applied original signal. From these windows BlackmanHPF provides high SQNR than other windowing techniques.Fig.10. FWHT applied Original signal filtered by FIR-HPF4.5 IIR FILTERINGThe Fig.9 and Fig.10 show the responses of FIR LPF and HPFdesigned by different windowing such as Hamming, Hanning,Blackman, Rectangular and Kaiser Window techniques of FWHTapplied original signal. From these windows Blackman HPFprovides high SQNR than other windowing techniques.The Fig.13 shows the response of IIR Filter designed byButterworth LPF, Butterworth HPF, Chebyshev LPF andChebyshev HPF of FFT applied original signal. From these resultsChebyshev HPF provides high SQNR than other techniques.484
ISSN: 2395-1680 (ONLINE)ICTACT JOURNAL ON MICROELECTRONICS, JANUARY 2018, VOLUME: 03, ISSUE: 04results Chebyshev HPF provides high SQNR than othertechniques.5. PERFORMANCE ANALYSISSimilarly, FIR and IIR filters are designed for each sub bandsseparately and their filter coefficients are quantized. Afterquantization the Signal to Quantization Noise Ratio for each filtercoefficients of original signal and their sub bands are calculatedand the comparison tables are made for the performance analysis.The signal to noise ratio value of signal after quantization is morecompared to signal to noise ratio value of signal beforequantization.Table.1. Table of SQNR for FFT applied signals filtered by FIR(LPF and HPF) Windowing methodsFig.13. FFT applied Original signal filtered by IIR arKaiserSignalsFilterOriginal Band1 Band2 Band3 Band4LPF 168.18 155.634 146.465 128.333 114.784HPF170.435 152.882 136.871 123.589 108.253LPF168.193 155.648 146.455 128.371 114.75HPF170.478 152.888 136.865 123.591 108.314LPF168.269 155.778 146.543 128.498 114.582HPF170.644 153.002 136.878 123.673 108.55LPF168.027 155.453 145.933 127.908 114.542HPF169.971 152.691 136.935 123.572 107.58LPF168.031 155.459 145.966 127.924 114.562HPF169.986 152.706 136.929 123.567 107.603From the Table.1 it inferred that, Blackman HPF windowprovides high SQNR (170.644) for original signal and BlackmanLPF window provides high SQNR for all four band signals(155.778 for Band1, 146.543 for Band2, 128.498 for Band3,114.582 for Band4 signals).Fig.14. FWHT applied Original signal filtered by IIR filterThe Fig.14 shows the response of IIR Filter designed byButterworth LPF, Butterworth HPF, Chebyshev LPF andChebyshev HPF of FWHT applied original signal. From theseresults Chebyshev HPF provides high SQNR than othertechniques.Table.2. Table of SQNR for FWHT applied signals filtered byFIR (LPF and HPF) Windowing larKaiserFig.15. DWT applied Original signal filtered by IIR filterSignalsFilterOriginal Band1 Band2 Band3 Band4LPF 168.18 155.635 146.468 128.333 114.785HPF170.426 152.878 136.877 123.585 108.254LPF168.194 155.649 146.458HPF170.467 152.885 136.871 123.587 108.315128.37 114.751LPF168.27HPF170.634 152.999 136.884 123.669 108.551LPF168.028 155.454 145.934 127.908 114.543HPF169.961 152.6929 136.941 123.567 107.582LPF168.032HPF169.977 152.7076 136.935 123.562 107.604155.779 146.546 128.498 114.584155.46145.967 127.924 114.564From the Table.2, it inferred that, Blackman HPF windowprovides high SQNR for original signal (170.634), Blackman LPFwindow provides high SQNR for Band1 signal (155.779), KaiserLPF window provides high SQNR for Band2 signal (145.967),Blackman LPF window provides high SQNR for Band3 signalThe Fig.15 shows the response of IIR Filter designed byButterworth LPF, Butterworth HPF, Chebyshev LPF andChebyshev HPF of DWT applied original signal. From these485
S SANGEETHA AND P KANNAN: DESIGN AND ANALYSIS OF DIGITAL FILTERS FOR SPEECH SIGNALS USING MULTIRATE SIGNAL PROCESSING(128.498) and Hamming LPF window provides high SQNR forBand4 signal (114.785).high SQNR in FIR filtering. In future, the proposed digital filterswill be further implement for various signals such as ECG signal,OFDM signal, etc.Table.3. Table of SQNR for DWT applied signals filtered byFIR (LPF and HPF) Windowing larKaiserREFERENCESSignalsFilter[1] Ankita Mahajan and Rajesh Mehra, “Design Analysis ofWord Length Effects in Interpolator Design”, InternationalJournal of Advanced Engineering Research andTechnology, Vol. 9, No. 1, pp. 104-108, 2014.[2] Atul Bhargava, “A Survey Report for Design of FIR Filterwith Different Method”, International Journal of Science,Engineering and Technology Research, Vol. 2, No. 2, pp.461-465, 2013.[3] Dolly Agrawal and Divya Kumud, “A Review On MultirateDigital Signal Processing”, International Journal ofElectrical and Electronics Engineers, Vol. 6, No. 2, pp. 324328, 2014.[4] Jagriti Saini and Rajesh Mehra, “Power Spectral DensityAnalysis of Speech Signal using Window Techniques”,International Journal of Computer Applications, Vol. 131,No. 14, pp. 33-36, 2015.[5] Lalima Singh, “Speech Signal Analysis using FFT andLPC”, International Journal of Advanced Research inComputer Engineering and Technology, Vol. 4, No. 4, pp.1658-1660, 2015.[6] Lalitha R. Naik and R.L. Devaraja Naik, “Sub-band Codingof Speech Signals using Multirate Signal Processing andComparing the various parameter of different SpeechSignals by Corrupting the same Speech Signal”,International Journal of Emerging Trends and Technologyin Computer Science, Vol. 4, No. 2, pp. 217-221, 2015.[7] Prajoy Podder, Tanvir Zaman Khan, Mamdudul HaqueKhan and M. Muktadir Rahman, “Comparative PerformanceAnalysis of Hamming, Hanning and Blackman Window”,International Journal of Computer Applications, Vol. 96,No. 18, pp. 1-7, 2014.[8] Rajeev Singh Dohare and Shilpa Datar, “QuantizedCoefficient FIR Filter for the design of Filter Bank”,International Journal of Engineering Trends andTechnology, Vol. 4, No. 8, pp. 13-18, 2013.[9] Samarjeet Singh and Uma Sharma, “ MATLAB basedDigital IIR Filter Design”, International Journal Electronicsand Computer Science Engineering, Vol. 1, No. 1, pp. 7483, 2014.[10] Saurabh Singh Rajput and S.S. Bhadauria, “Implementationof FIR Filter using Efficient Window Function and ItsApplication In Filtering a Speech Signal”, InternationalJournal of Electrical, Electronic and Mechanical Controls,Vol. 1, No. 2, pp. 158-164, 2012.[11] Suraj R. Gaikwad and Gopal Gawande, “Review: Design ofHighly Efficient Multirate Digital Filters”, InternationalJournal of Engineering Research and Applications, Vol. 3,No. 6, pp. 560-564, 2013.[12] Suverna Sengar and Partha Pratim Bhattacharya, ‘MultirateFiltering for Digital Signal Processing and its Applications’,ARPN Journal of Science and Technology, Vol. 2, No. 3, pp.228-237, 2012.[13] Suraj R. Gaikwad and Gopal S. Gawande, “Implementationof Efficient Multirate Filter Structure for Decimation”,Original Band1 Band2 Band3 Band4LPF 168.279 155.862 146.548 128.549 114.981HPF170.802 153.681 137.368 124.654 108.626LPF168.293155.88HPF170.84153.711 137.358 124.621 108.684146.556 128.587 114.939LPF168.369 156.017 146.617 128.724 114.814HPF171.023 153.772 137.356LPF168.115 155.663 145.852 128.109 114.675HPF170.277 161.246 137.481LPF168.119 155.669 145.885 128.126 114.696HPF170.295 162.391 137.474 124.785 108.017124.61 108.911124.79 107.996From the Table.3 it inferred that, Blackman HPF windowprovides high SQNR for original signal (171.023), Kaiser HPFwindow provides high SQNR for Band1 signal (162.391),Hanning LPF window provides high SQNR for Band2 signal(146.556), Blackman LPF window provides high SQNR forBand3 signal (128.724) and Hamming LPF window provides highSQNR for Band4 signal (114.981).Table.4. Table of SQNR for different transforms applied signalsfiltered by IIR (LPF and KaiserFilterSignalsLPFOriginal Band1166.795 152.43HPF166.794 152.431 138.433 124.218 109.091152.97Band2 Band3 Band4138.431 124.217 109.092LPF166.994HPF174.106 156.623 139.207 124.123 113.47156.62139.061 124.958 109.928LPF174.104HPF174.649 157.169 139.979 125.129 115.333139.208 124.119 113.472LPF166.776 152.819 138.324 124.836 107.89HPF166.778 152.8196 138.324 124.838 107.89LPF166.938 153.085 139.128 125.541 109.134HPF174.353 156.484138.76124.073 114.038From the Table.4, it inferred that Chebyshev HPF provideshigh SQNR for original signal and Band4 signal, ButterworthHPF provides high SQNR for Band1 and Band2 signals andChebyshev LPF provide high SQNR for Band3 signal.6. CONCLUSIONThe proposed digital filter for speech signals using multi ratesignal processing has been designed. After filtering, thequantization of the input signal and filter coefficients wereanalyzed. The performance of the proposed system was evaluatedbased on Signal to Quantization Noise Ratio. From theperformance measures, it was observed that DWT applied signalsfiltered by Chebyshev HPF provides high SQNR in IIR Filteringand DWT applied signals filtered by Blackman Window provide486
ISSN: 2395-1680 (ONLINE)ICTACT JOURNAL ON MICROELECTRONICS, JANUARY 2018, VOLUME: 03, ISSUE: 04[16] Vishv Mohan, “Analysis and Synthesis of Speech usingMatlab”, International Journal of Advancements inResearch and Technology, Vol. 2, No. 5, pp. 373-382, 2013.[17] Vaibhav M Dikhole et al, “Effect of Finite Word Length forFIR Filter Coefficient in Electrocardiogram Filtering”,Proceedings of National Conference on Innovative Trendsin Science and Engineering, Vol. 4, No. 7, pp. 266-271,2016.International Journal of Current Engineering andTechnology, Vol. 4, No. 2, pp. 1008-1010, 2014.[14] P. Suresh Babu, D. Srinivasulu Reddy and P.V.N. Reddy,“Speech Signal Analysis using Windowing Techniques”,International Journal of Emerging Trends in EngineeringResearch, Vol. 3, No. 6, pp. 257-263, 2015.[15] Vijayakumar Majjagi, “Sub Band Coding of Speech Signalby using Multi-Rate Signal Processing”, InternationalJournal of Engineering Research and Technology, Vol. 2,No. 9, pp. 45-49, 2013.487
3.3 FILTERING TECHNIQUES Filtering techniques are useful for separating, extracting and restoring the signal. It helps to remove the sources of noise. There are two types of digital filtering techniques. FIR filtering technique IIR filtering technique 3.3.1 FIR Filter Design: FIR Filter the presentoutput depends on input and past input.
