GENERAL KALMAN FILTER & SPEECH ENHANCEMENT FOR SPEAKER IDENTIFICATION

Transcription

1 International Journal on Cybernetics & Informatics (IJCI) Vol 5, No 4, August 26 ABSRAC GENERAL KALMAN FILER & SPEECH ENHANCEMEN FOR SPEAKER IDENIFICAION Vijay Kiran Battula and Appala Naidu Gottapu Department of ECE, JNUK-UCEV, Vizianagaram, Andhra Pradesh, India Presence of noise increases the dimension of the information A noise suppression algorithm is deeloped with an idea of combining the General Kalman Filter and Estimate Maximization (EM) frame worhis combination is helpful and effectie in identifying noise characteristics of an acoustic enironment Recursion between Estimate step and Maximization step enabled the algorithm to deal any model of noise he same Speech enhancement procedure in applied in the pre-processing stage of a conentional Speaer identification method Due to the non-stationary nature of noise and speech adaptie algorithms are required Algorithm is first applied for Speech enhancement problem and then extended to using it in the pre-processing step of the Speaer identification he present wor is compared in terms of significant metrics with existing and popular algorithms and results show that the deeloped algorithm is dominant oer them KEYWORDS Speech processing, Speech enhancement, Speaer identification, General Kalman filter and EM algorithm INRODUCION he two major applications of the speech processing namely speech enhancement and speaer identification are inter related in terms of the core techniques used for performing them Speech enhancement directly resembles the pre-processing stage of the speaer identification procedure in most of the cases he wor depicts the relation and significance of deeloping an algorithm to address the core areas of their respectie step wise procedure is the area of interest Speech and noise in a natural enironment always finds in a combined form, Speech is eery time degraded by the noise General assumption is to always find an addiction of Gaussian noise to the speech information signal his noise aries in its characteristics for different acoustic enironments he idea is to deelop a robust algorithm to deal a noise of any enironment that is degrading speech Elimination of noise could be een possible through the transducers lie microphone used to record the speech information and conert them to the electrical equialent representation here are techniques deeloped based on the number of microphones to be used DOI: 52/ijci

2 International Journal on Cybernetics & Informatics (IJCI) Vol 5, No 4, August 26 to identify the noise components in the signal he present wor is based on Single microphone speech enhancement Noise in the problem is first modelled as an Auto Regressie process his performed as in [3] Estimate Maximization frame wor is studied which inoles speech enhancement and as well as parameter estimation General Kalman Filter(GKF) is a time domain algorithm first studied for using it system identification problem lie echo cancellation in [4] In Echo cancellation echo path estimation is performed using GKF For extending the General Kalman filter to the speech enhancement it has to be gien with at least one among the nowledge of two characteristics of speech or noise EM frame wor helps in proiding the noise characteristics he time domain implementation of the Kalman filter is difficult and not dominant GEnzner through his studies proposed the effectie implementation of Kalman filter in frequency domain [2] But GKF is time domain implemented ersion of Kalman filter and hence draws the attention Speaer Recognition is a part behaioural characteristic of the Biometrics he two major classifications are Speaer erification and Speaer identification namely Speaer identification can be either text dependent or text independent In this paper a text independent speaer identification method is considered with a pre-processing stage of deeloped algorithm 2 DEFINIION OF HE PROBLEM AND RELAED CONCEPS x( is a noisy speech signal, s( and ( are its two components x ( = s( + ( Clean speech signal is represented with s( and it obtained by supressing additie noise ( in x ( 2 Aim Single microphone is used to record the degraded speech he deeloped algorithm is aimed to use in studying then noise characteristics and suppression of the noise components from the signal hus to proide the speech quality enhancement and intelligibility 22 General Kalman Filter In [4] General Kalman filter is implemented with two estimation problems he first is a hidden Maro modelled impulse response coefficient estimation and the second is the estimation of desired response with nowledge of the aboe estimated coefficients and far end signal In this wor the problem is redefined to fit for a speech enhancement problem he similar coefficient estimation is equialent to the change affecting the noise characteristics of the acoustic enironment under consideration And these are updated for find new noise model from preious effects as recursie process by applying estimate maximization frame wor 8

3 International Journal on Cybernetics & Informatics (IJCI) Vol 5, No 4, August Estimate Maximization frame wor here are two steps in this process namely estimation step and maximization step It is a recursie procedure And the ending if the recursion is predefined or left to the intelligibility of the algorithm General Kalman filter proides estimation of the required entity in estimation step hese parameters inoled in the estimation are identified and updated in the maximization or parameter estimation step and again fed to estimation step his repeats as an iteratie process until the clean entity of interest is obtained 24 Proposed Methods he Figure represent the bloc diagram of combined speech enhancement for speaer identification, the first half of the diagram illustrates the process flow of speech enhancement as follows i Noise degraded speech information is collected through single microphone ii his is gien to the speech enhancement stage iii he first step of speech enhancement contains the segmentation of speech signal satisfying the stationary condition of the signal i hus frames are extracted from the source information and each frame is now forwarded for noise suppression indiidually he AR model of the noise of the acoustic enironment is acquired prior to the aboe steps and state space formulation is done to it i he estimation process starts by giing noise estimate and present speech frame to General Kalman filter to find optimum output ii he parameters are updated in maximization step and again gien to GKF for new estimation iii his Process continues until clean speech frame is obtained for a fixed number of iterations ix Finally all the frames are concatenated to obtain clean speech 9

4 International Journal on Cybernetics & Informatics (IJCI) Vol 5, No 4, August 26 Figure Proposed Speech Enhancement for Speaer identification Figure represents the proposed method for Speaer identification i he initial stage of speaer identifications always be the speech enhancement stage he output of the Speech enhancement stage is a clean speech ii raining phase and Verification are two stages of identification In training phase speech features of the concerned group are collected and stored in data base iii In erification stage they are retrieed to compare with the features of persons speech waiting for authentication i If the matching criteria satisfies authentication is proided otherwise rejected In this paper Eigen features based speech features are extracted in feature extraction stage MFCC range of frequencies is under consideration i Dimensionality reduction techniques lie principal component analysis (PCA) and Independent Component analysis (ICA) used for reducing the feature space ii Genetic Algorithm is used for erification procedure to proide optimum results and the cost function of it is defined below L L L ( ) ( ) 2 2 max ( m) + max P( m, m= R λ () m= m= n+ R=Auto correlation (Correlation between each speech feature itself), P=Cross Correlation (Correlation between speech feature with other speech feature),l=length of each speech feature,weighing factor λ 2

5 International Journal on Cybernetics & Informatics (IJCI) Vol 5, No 4, August 26 3 ALGORIHM 3 Auto Regressie modelling of noise Let us consider the problem in a discrete time index of n, x ( = s( + ( (2) Noise and speech are non-stationary Hence, any one of them has to be nown in order to go for formulation Consider the speech absent periods to study noise characteristics, x ( = ( (3) Where ( n ) is signal collected by recording deice when speech is absent and use as initial signal ector of noise Noise is modelled as stochastic AR process: g ( = λ q( n ) + u ( (4) 2 q= λ, λ 2, λ represents AR parameters of noise process and normalized (zero-mean unit ariance),white Gaussian noise g represents power leel u ( is 32 E-step AR model of noise process is conerted into state space formulation + Φ ( = [ ( n + ), ( n 2),, ( ] = λ λ λ 2 λ 3 λ 2 dimensional ectors g [ g ] is noise transition matrix and - λ = he below equation proides estimate of noise 2

6 33 M-step International Journal on Cybernetics & Informatics (IJCI) Vol 5, No 4, August 26 ( = Φ ( n ) g ( u ( (5) + x( = s( + h ( ( (6) s( = x( h ( ( (7) Parameters for estimate and AR model, GKF algorithm procedural steps results in finding the minimal error and optimal estimate he following are the related equations Initialize with h ˆ ( = and R ( ) εi q where ε is small positie constant µ = hˆ ( = hˆ ( n ) w( (8) + x( = s( + h ( ( (9) e( = x( ˆ( () 2 R m ( = Rµ ( n ) + σ w ( I q () e R ( = ( R ( ( + σ ( I (2) m 2 p K ( = R ( ( R e ( (3) m e( = x( ( hˆ ( n ) (4) hˆ ( = hˆ ( n ) K( e( (5) + Where R m ( is priori misalignment, ( R ( = µ [ I K( ( ] R ( (6) R µ q m is posteriori misalignment correlation matrix, K ( is Kalman gain, I q is identity matrix, σ 2 ( n w ) is ariance of w (, R e ( is priori error ector correlation matrix and e ( is error between signal and estimated Let θ be the ector of unnown parameters and gien as ector θ = [ λ g ] 22

7 International Journal on Cybernetics & Informatics (IJCI) Vol 5, No 4, August 26 Z = q ( n ) q ( n ) (7) Where updated parameters are defined as 4 RESULS 4 Simulation and Metrics ˆ ( q+ ) Y = q ( n ) ( (8) λ Y (9) = E[ Z ] ( ˆ( q+ ) ( q+ ) = ) gˆ Z + E[ Y ] λ (2) NOIZEUS speech corpus contains noisy speech signals hese are directly used to test the performance of the proposed algorithm he similar experiments are conducted with popular algorithms and are used for comparison Pea signal to Noise ratio and Mean square error are used to ealuate performance for Speech enhancement Identification rate and elapsed time for a fixed number of generations are compared for Speaer identification Simulations are performed in MALAB platform and respectie analysis is presented below Figure 2 Noisy and GKF enhanced babble noise type of SNR db, 5dB, 5dB 23

8 International Journal on Cybernetics & Informatics (IJCI) Vol 5, No 4, August 26 Figure 3PSNR and MSE comparison of SS, NLMS, RLS, APA and GKF (Speech Enhancement) 24

9 International Journal on Cybernetics & Informatics (IJCI) Vol 5, No 4, August 26 Figure 4 Identification rate comparison of different processing method combinations (Speaer Identificatio 5 CONCLUSIONS General Kalman filter resembles Affine projection algorithm (APA) for certain considerations but the combination with Estimate Maximization is more efficient for GKF Figure 2 presents the MALAB simulated results of babble noise type of arious SNR leels Figure 3 is the analysis of proposed algorithm, Normalized least mean square algorithm, Recursie least squares, APA and Spectral subtraction for Speech enhancement applications Proposed algorithm has gien better results in terms of Mean square error and Pea signal to noise ratio Figure 4 clearly shows the dimensionality reduction effect on the speaer identification system with change in filtration scheme he respectie identification rates and elapsed time of different methods and combination under considerations presented in figure 4 shows GKF is optimum his wor could be further extended to real time and then performance has to analyzed to now its compatibility REFERENCES [] R E Kalman, A new approach to linear filtering and prediction problems, J Basic Eng, ol82, pp 35 45, Mar 96 [2] G Enzner and P Vary, Frequency-domain adaptie Kalman filter foracoustic echo control in handsfree telephones, Signal Process, ol86, pp 4 56, 26 [3] Jae Lim; A Oppenheim, All-pole modeling of degraded speech, IEEE ransactions on Acoustics, Speech, and Signal ProcessingYear: 978, Volume: 26, Issue: 3 [4] C Paleologu, J Benesty, and S Ciochina, Study of the general Kalman filter for echo cancellation, IEEE rans Audio, Speech, LanguageProcessing, ol 2, pp , Aug 23 25

10 International Journal on Cybernetics & Informatics (IJCI) Vol 5, No 4, August 26 [5] S E Bou-Ghazale and K Assaleh, A robust endpoint detection of speech for noisy enironments with application to automatic speech recognition, in Proc ICASSP22, ol 4, pp , 22 [6] S Gannot, Speech processing utilizing the Kalman filter, IEEE Instrum Meas Mag, ol 5, no 3, pp 4, Jun 22 [7] Kishore Kumar, Sia Prasad Nandyala, Rai Bolimera, Speech Enhancement using Spectral Subtraction,Affine Projection Algorithms and Classical AdaptieFilters, International Conference On Adances In Electronics, Electrical And Computer Science Engineering - EEC 22 [8] Sadaoi Furui, 5 Years of Progress in Speech and Speaer Recognition Research, Ecti ransactions on computer and Information echnology, Vol, No2, Noember 25 [9] Omar Daoud, Abdel-Rahman Al-Qawasmi and Khaled daqrouq, Modified PCA Speaer Identification Based System Using Waelet ransform and Neural Networs, International Journal of Recent rends in Engineering, Vol 2, No 5, Noember 29 [] Reynolds, DA, Experimental ealuation of features for robust speaer identification, IEEE ransactions on SAP, Vol 2, pp , 28 [] K Koteswara rao, G Appala Naidu, Eigen features based speaer identification model using Genetic algorithm, VSPICE-25 National conference, JNU KAKINADA AUHORS BVijay Kiran receied Bech degree from JNUK and currently pursuing Mech degree in UCEV Vizianagaram, JN Uniersity Kainada and Vizianagaram G Appala Naidu receied both Bech and Mech degrees from JNUH He is currently woring in Uniersity College of Engineering Vizianagaram, JN Uniersity Kainada, and Vizianagaram 26