Research article

Improving speech recognition using bionic wavelet features

  • Bionic wavelet transform is a continuous wavelet, based on adaptive time frequency technique. This paper presents a speech recognition system for recognizing isolated words by discretizing the continuous Bionic Wavelet (BW). Conversion from continuous to discrete is achieved by adopting central frequency and thresholding techniques. The BW features of noisy signal are processed through MFCC to obtain the optimal features of the speech signal. SVM, Artificial Neural Network (ANN) and LSTM techniques are used to improve the recognition rate by enhancing the speech signals. The experiments are conducted on FSDD and Kannada data set. The speech feature vector is calculated using the parameters extracted by Bionic wavelet with different central frequencies of Morlet, Daubechies and Bior3.5, coiflet5 mother wavelets. The obtained Bionic-MFCC optimal features are fed to SVM, ANN and LSTM models for the classification and recognition process. The performance of the models is tabulated for correct recognition that varies from 95% to 96% among these models. The models are tested for various SNRs noise levels like 5 dB, 10 dB, 15 dB and the recognition accuracies of these models are presented for convoluted noisy speech data.

    Citation: Vani H Y, Anusuya M A. Improving speech recognition using bionic wavelet features[J]. AIMS Electronics and Electrical Engineering, 2020, 4(2): 200-215. doi: 10.3934/ElectrEng.2020.2.200

    Related Papers:

    [1] Zhiyuan Wang, Chu Zhang, Shaopei Xue, Yinjie Luo, Jun Chen, Wei Wang, Xingchen Yan . Dynamic coordinated strategy for parking guidance in a mixed driving parking lot involving human-driven and autonomous vehicles. Electronic Research Archive, 2024, 32(1): 523-550. doi: 10.3934/era.2024026
    [2] Xiaoying Zheng, Jing Wu, Xiaofeng Li, Junjie Huang . UAV search coverage under priority of important targets based on multi-location domain decomposition. Electronic Research Archive, 2024, 32(4): 2491-2513. doi: 10.3934/era.2024115
    [3] Yu Shen, Hecheng Li . A multi-strategy genetic algorithm for solving multi-point dynamic aggregation problems with priority relationships of tasks. Electronic Research Archive, 2024, 32(1): 445-472. doi: 10.3934/era.2024022
    [4] Sida Lin, Lixia Meng, Jinlong Yuan, Changzhi Wu, An Li, Chongyang Liu, Jun Xie . Sequential adaptive switching time optimization technique for maximum hands-off control problems. Electronic Research Archive, 2024, 32(4): 2229-2250. doi: 10.3934/era.2024101
    [5] Ismail Ben Abdallah, Yassine Bouteraa, Saleh Mobayen, Omar Kahouli, Ali Aloui, Mouldi Ben Amara, Maher JEBALI . Fuzzy logic-based vehicle safety estimation using V2V communications and on-board embedded ROS-based architecture for safe traffic management system in hail city. Electronic Research Archive, 2023, 31(8): 5083-5103. doi: 10.3934/era.2023260
    [6] Jian Gong, Yuan Zhao, Jinde Cao, Wei Huang . Platoon-based collision-free control for connected and automated vehicles at non-signalized intersections. Electronic Research Archive, 2023, 31(4): 2149-2174. doi: 10.3934/era.2023111
    [7] Hao Li, Zhengwu Wang, Shuiwang Chen, Weiyao Xu, Lu Hu, Shuai Huang . Integrated optimization of planning and operation of a shared automated electric vehicle system considering the trip selection and opportunity cost. Electronic Research Archive, 2024, 32(1): 41-71. doi: 10.3934/era.2024003
    [8] Wenjie Wang, Suzhen Wen, Shen Gao, Pengyi Lin . A multi-objective dynamic vehicle routing optimization for fresh product distribution: A case study of Shenzhen. Electronic Research Archive, 2024, 32(4): 2897-2920. doi: 10.3934/era.2024132
    [9] Yineng Ouyang, Zhaotao Liang, Zhihui Ma, Lei Wang, Zhaohua Gong, Jun Xie, Kuikui Gao . A class of constrained optimal control problems arising in an immunotherapy cancer remission process. Electronic Research Archive, 2024, 32(10): 5868-5888. doi: 10.3934/era.2024271
    [10] Michael Barg, Amanda Mangum . Statistical analysis of numerical solutions to constrained phase separation problems. Electronic Research Archive, 2023, 31(1): 229-250. doi: 10.3934/era.2023012
  • Bionic wavelet transform is a continuous wavelet, based on adaptive time frequency technique. This paper presents a speech recognition system for recognizing isolated words by discretizing the continuous Bionic Wavelet (BW). Conversion from continuous to discrete is achieved by adopting central frequency and thresholding techniques. The BW features of noisy signal are processed through MFCC to obtain the optimal features of the speech signal. SVM, Artificial Neural Network (ANN) and LSTM techniques are used to improve the recognition rate by enhancing the speech signals. The experiments are conducted on FSDD and Kannada data set. The speech feature vector is calculated using the parameters extracted by Bionic wavelet with different central frequencies of Morlet, Daubechies and Bior3.5, coiflet5 mother wavelets. The obtained Bionic-MFCC optimal features are fed to SVM, ANN and LSTM models for the classification and recognition process. The performance of the models is tabulated for correct recognition that varies from 95% to 96% among these models. The models are tested for various SNRs noise levels like 5 dB, 10 dB, 15 dB and the recognition accuracies of these models are presented for convoluted noisy speech data.


    We consider the system of Hamilton-Jacobi equations

    {λu1(x)+H1(Du1(x))+B1(u1(x),u2(x))=0  in Tn,λu2(x)+H2(Du2(x))+B2(u1(x),u2(x))=0  in Tn, (1.1)

    where λ>0 is a given constant, the functions Hi:RnR and Bi:R2R, with i=1,2, are given continuous functions, and Tn denotes the n-dimensional flat torus Rn/Zn.

    In a recent paper [6], the authors have investigated the vanishing discount problem for a nonlinear monotone system of Hamilton-Jacobi equations

    {λu1(x)+G1(x,Du1(x),u1(x),u2(x),,um(x))=0  in Tn,λu1(x)+G1(x,Du1(x),u1(x),u2(x)λum(x)+Gm(x,Dum(x),u1(x),u2(x),,um(x))=0  in Tn, (1.2)

    and established under some hypotheses on the GiC(Tn×Rn×Rm) that, when uλ=(uλ,1,,uλ,m)C(Tn)m denoting the (viscosity) solution of (1.2), the whole family {uλ}λ>0 converges in C(Tn)m to some u0C(Tn)m as λ0+. The constant λ>0 in the above system is the so-called discount factor.

    The hypotheses on the system are the convexity, coercivity, and monotonicity of the Gi as well as the solvability of (1.2), with λ=0. Here the convexity of Gi is meant that the functions Rn×Rm(p,u)Gi(x,p,u) are convex. We refer to [6] for the precise statement of the hypotheses.

    Prior to work [6], there have been many contributions to the question about the whole family convergence (in other words, the full convergence) under the vanishing discount, which we refer to [1,3,4,6,8,9,10] and the references therein.

    In the case of the scalar equation, B. Ziliotto [11] has recently shown an example of the Hamilton-Jacobi equation having non-convex Hamiltonian in the gradient variable for which the full convergence does not hold. In Ziliotto's approach, the first step is to find a system of two algebraic equations

    {λu+f(uv)=0,λv+g(vu)=0, (1.3)

    with two unknowns u,vR and with a parameter λ>0 as the discount factor, for which the solutions (uλ,vλ) stay bounded and fail to fully converge as λ0+. Here, an "algebraic" equation is meant not to be a functional equation. The second step is to interpolate the two values uλ and vλ to get a function of xT1 which satisfies a scalar non-convex Hamilton-Jacobi equation in T1.

    In the first step above, Ziliotto constructs f,g based on a game-theoretical and computational argument, and the formula for f,g is of the minimax type and not quite explicit. In [5], the author has reexamined the system given by Ziliotto, with a slight generality, as a counterexample for the full convergence in the vanishing discount.

    Our purpose in this paper is to present a system (1.3), with an explicit formula for f,g, for which the solution (uλ,vλ) does not fully converge to a single point in R2. A straightforward consequence is that (1.1), with B1(u1,u2)=f(u1u2) and B2(u1,u2)=g(u2u1), has a solution given by

    (uλ,1(x),uλ.2(x))=(uλ,vλ)   for xTn,

    under the assumption that Hi(x,0)=0 for all xTn, and therefore, gives an example of a discounted system of Hamilton-Jacobi equations, the solution of which fails to satisfy the full convergence as the discount factor goes to zero.

    The paper consists of two sections. This introduction is followed by Section 2, the final section, which is divided into three subsections. The main results are stated in the first subsection of Section 2, the functions f,g, the key elements of (1.3), are contstructed in the second subsection, and the final subsection provides the proof of the main results.

    Our main focus is now the system

    {λu+f(uv)=0λv+g(vu)=0, (2.1)

    where f,gC(R,R) are nondecreasing functions, to be constructed, and λ>0 is a constant, to be sent to zero. Notice that (2.1) above is referred as (1.3) in the previous section.

    We remark that, due to the monotonicity assumption on f,g, the mapping (u,v)(f(uv),g(vu)),R2R2 is monotone. Recall that, by definition, a mapping (u,v)(B1(u,v),B2(u,v)),R2R2 is monotone if, whenever (u1,v1),(u2,v2)R2 satisfy u1u2v1v2 (resp., v1v2u1u2), we have B1(u1,v1)B1(u2,v2) (resp., B2(u1,v1)B2(u2,v2)).

    Our main results are stated as follows.

    Theorem 1. There exist two increasing functions f,gC(R,R) having the properties (a)–(c):

    (a) For any λ>0 there exists a unique solution (uλ,vλ)R2 to (2.1),

    (b) the family of the solutions (uλ,vλ) to (2.1), with λ>0, is bounded in R2,

    (c) the family {(uλ,vλ)}λ>0 does not converge as λ0+.

    It should be noted that, as mentioned in the introduction, the above theorem has been somewhat implicitly established by Ziliotto [11]. In this note, we are interested in a simple and easy approach to finding functions f,g having the properties (a)–(c) in Theorem 1.

    The following is an immediate consequence of the above theorem.

    Corollary 2. Let HiC(Rn,R), i=1,2, satisfy H1(0)=H2(0)=0. Let f,gC(R,R) be the functions given by Theorem 1, and set B1(u1,u2)=f(u1u2) and B2(u1,u2)=g(u2u1) for all (u1,u2)R2. For any λ>0, let (uλ,1,uλ,2) be the (viscosity) solution of (1.1). Then, the functions uλ,i are constants, the family of the points (uλ,1,uλ,2) in R2 is bounded, and it does not converge as λ0+.

    Notice that the convexity of Hi in the above corollary is irrelevant, and, for example, one may take Hi(p)=|p|2 for iI, which are convex functions.

    We remark that a claim similar to Corollary 2 is valid when one replaces Hi(p) by degenerate elliptic operators Fi(x,p,M) as far as Fi(x,0,0)=0, where M is the variable corresponding to the Hessian matrices of unknown functions. (See [2] for an overview on the viscosity solution approach to fully nonlinear degenerate elliptic equations.)

    If f,g are given and (u,v)R2 is a solution of (2.1), then w:=uv satisfies

    λw+f(w)g(w)=0. (2.2)

    Set

    h(r)=f(r)g(r)   for rR, (2.3)

    which defines a continuous and nondecreasing function on R.

    To build a triple of functions f,g,h, we need to find two of them in view of the relation (2.3). We begin by defining function h.

    For this, we discuss a simple geometry on xy-plane as depicted in Figure 1 below. Fix 0<k1<k2. The line y=12k2+k1(x+12) has slope k1 and crosses the lines x=1 and y=k2x at P:=(1,12(k1+k2)) and Q:=(12,12k2), respectively, while the line y=k2x meets the lines x=1 and x=12 at R:=(1,k2) and Q=(12,12k2), respectively.

    Figure 1.  Graph of ψ.

    Choose k>0 so that 12(k1+k2)<k<k2. The line y=kx crosses the line y=12k2+k1(x+12) at a point S:=(x,y) in the open line segment between the points P=(12,12(k1+k2)) and Q=(12,12k2). The line connecting R=(1,k2) and S=(x,y) can be represented by y=k2+k+(x+1), with k+:=y+k2x+1>k2.

    We set

    ψ(x)={k2x for x(,1][1/2,),min{k2+k+(x+1),12k2+k1(x+12)}   for x(1,12).

    It is clear that ψC(R) and increasing on R. The building blocks of the graph y=ψ(x) are three lines whose slopes are k1<k2<k+. Hence, if x1>x2, then ψ(x1)ψ(x2)k1(x1x2), that is, the function xψ(x)k1x is nondecreasing on R.

    Next, we set for jN,

    ψj(x)=2jψ(2jx)   for xR.

    It is clear that for all jN, ψjC(R), the function xψj(x)k1x is nondecreasing on R, and

    ψj(x){>k2x   for all x(2j,2j1),=k2x   otherwise.

    We set

    η(x)=maxjNψj(x)   for xR.

    It is clear that ηC(R) and xη(x)k1x is nondecreasing on R. Moreover, we see that

    η(x)=k2x   for all x(,12][0,),

    and that if 2j<x<2j1 and jN,

    η(x)=ψj(x)>k2x.

    Note that the point S=(x,y) is on the graph y=ψ(x) and, hence, that for any jN, the point (2jx,2jy) is on the graph y=η(x). Similarly, since the point S=(x,y) is on the graph y=kx and for any jN, the point (2jx,2jy) is on the graph y=kx. Also, for any jN, the point (2j,k22j) lies on the graphs y=η(x) and y=k2x.

    Fix any d1 and define hC(R) by

    h(x)=η(xd).

    For the function h defined above, we consider the problem

    λz+h(z)=0. (2.4)

    Lemma 3. For any λ0, there exists a unique solution zλR of (2.4).

    Proof. Fix λ0. The function xh(x)+λx is increasing on R and satisfies

    limx(h(x)+λx)=   and   limx(h(x)+λx)=.

    Hence, there is a unique solution of (2.4).

    For any λ0, we denote by zλ the unique solution of (2.4). Since h(d)=0, it is clear that z0=d.

    For later use, observe that if λ>0, k>0, and (z,w)R2 is the point of the intersection of two lines y=λx and y=k(xd), then w=λz=k(zd) and

    z=kdk+λ. (2.5)

    Lemma 4. There are sequences {μj} and {νj} of positive numbers converging to zero such that

    zμj=k2dk2+μj  and  zνj=kdk+νj.

    Proof. Let jN. Since (2j,k22j) is on the intersection of the graphs y=k2x and y=η(x), it follows that (2j+d,k22j) is on the intersection of the graphs y=k2(xd) and y=h(x). Set

    μj=k22jd2j, (2.6)

    and note that μj>0 and that

    μj(d2j)=k22j,

    which says that the point (d2j,k22j) is on the line y=μjx. Combining the above with

    k22j=h(d2j)

    shows that d2j is the unique solution of (2.4). Also, since (d2j,μj(d2j))=(d2j,k22j) is on the line y=k2(xd), we find by (2.5) that

    zμj=k2dk2+μj.

    Similarly, since (2jx,2jy) is on the intersection of the graphs y=kx and y=η(x), we deduce that if we set

    νj:=2jyd+2jx=2j|y|d2j|x|, (2.7)

    then

    zνj=kdk+νj.

    It is obvious by (2.6) and (2.7) that the sequences {μj}jN and {νj}jN are decreasing and converge to zero.

    We fix k0(0,k1) and define f,gC(R) by f(x)=k0(xd) and

    g(x)=f(x)h(x).

    It is easily checked that g(x)(k1k0)x is nondecreasing on R, which implies that g is increasing on R, and that h(x)=f(x)g(x) for all xR. We note that

    f(d)=h(d)=g(d)=0. (2.8)

    We fix f,g,h as above, and consider the system (2.1).

    Lemma 5. Let λ>0. There exists a unique solution of (2.1).

    The validity of the above lemma is well-known, but for the reader's convenience, we provide a proof of the lemma above.

    Proof. By choice of f,g, the functions f,g are nondecreasing on R. We show first the comparison claim: if (u1,v1),(u2,v2)R2 satisfy

    λu1+f(u1v1)0,λv1+g(v1u1)0, (2.9)
    λu2+f(u2v2)0,λv2+g(v2u2)0, (2.10)

    then u1u2 and v1v2. Indeed, contrary to this, we suppose that max{u1u2,v1v2}>0. For instance, if max{u1u2,v1v2}=u1u2, then we have u1v1u2v2 and u1>u2, and moreover

    0λu1+f(u1v1)λu1+f(u2v2)>λu2+f(u2v2),

    yielding a contradiction. The other case when max{u1u2,v1v2}=v1v2, we find a contradiction, 0>λv2+g(v2u2), proving the comparison.

    From the comparison claim, the uniqueness of the solutions of (2.1) follows readily.

    Next, we may choose a constant C>0 so large that (u1,v1)=(C,C) and (u2,v2)=(C,C) satisfy (2.9) and (2.10), respectively. We write S for the set of all (u1,u2)R2 such that (2.9) hold. Note that (C,C)S and that for any (u,v)S, uC and vC. We set

    u=sup{u:(u,v)S  for some v},v=sup{v:(u,v)S  for some u}.

    It follows that Cu,vC. We can choose sequences

    {(u1n,v1n)}nN,{(u2n,v2n)}nNS

    such that {u1n},{v2n} are nondecreasing,

    limnu1n=u   and   limnv2n=v.

    Observe that for all nN, u2nu, v1nv, and

    0λu1n+f(u1nv1n)λu1n+f(u1nv),

    which yields, in the limit as n,

    0λu+f(uv).

    Similarly, we obtain 0λv+g(vu). Hence, we find that (u,v)S.

    We claim that (u,v) is a solution of (2.1). Otherwise, we have

    0>λu+f(uv)   or   0>λv+g(vu).

    For instance, if the former of the above inequalities holds, we can choose ε>0, by the continuity of f, so that

    0>λ(u+ε)+f(u+εv).

    Since (u,v)S, we have

    0λv+g(vu)λv+g(vuε).

    Accordingly, we find that (u+ε,v)S, which contradicts the definition of u. Similarly, if 0>λv+g(vu), then we can choose δ>0 so that (u,v+δ)S, which is a contradiction. Thus, we conclude that (u,v) is a solution of (2.1).

    Theorem 6. For any λ>0, let (uλ,vλ) denote the unique solution of (2.1). Let {μj},{νj} be the sequences of positive numbers from Lemma 2.4. Then

    limjuμj=k0dk2  and  limjuνj=k0dk.

    In particular,

    lim infλ0uλk0dk2<k0dklim supλ0uλ.

    With our choice of f,g, the family of solutions (uλ,vλ) of (2.1), with λ>0, does not converge as λ0.

    Proof. If we set zλ=uλvλ, then zλ satisfies (2.4). By Lemma 4, we find that

    zμj=k2dk2+μj   and   zνj=kdk+νj.

    Since uλ satisfies

    0=λuλ+f(zλ)=λuλ+k0(zλd),

    we find that

    uμj=k0(zμjd)μj=k0dμj(k2k2+μj1)=k0dμjμjk2+μj=k0dk2+μj,

    which shows that

    limjuμj=k0dk2.

    A parallel computation shows that

    limjuνj=k0dk.

    Recalling that 0<k<k2, we conclude that

    lim infλ0uλk0dk2<k0dklim supλ0uλ.

    We remark that, since

    limλ0zλ=d   and   vλ=uλzλ,
    limjvμj=k0dk2d   and   limjvνj=k0dkd.

    We give the proof of Theorem 1.

    Proof of Theorem 1. Assertions (a) and (c) are consequences of Lemma 5 and Theorem 6, respectively.

    Recall (2.8). That is, we have f(d)=h(d)=g(d)=0. Setting (u2,v2)=(d,0), we compute that for any λ>0,

    λu2+f(u2v2)>f(d)=0   and   λv2+g(v2u2)=g(d)=0.

    By the comparison claim, proved in the proof of Lemma 5, we find that uλd and vλ0 for any λ>0. Similarly, setting (u1,v1)=(0,d), we find that for any λ>0,

    λu1+f(u1v1)=f(d)=0   and   λv1+g(v1u1)g(v1u1)=g(d)=0,

    which shows by the comparison claim that uλ0 and vλd for any λ>0. Thus, the sequence {(uλ,vλ)}λ>0 is bounded in R2, which proves assertion (b).

    Proof of Corollary 2. For any λ>0, let (uλ,vλ)R2 be the unique solution of (2.1). Since H1(0)=H2(0)=0, it is clear that the constant function (uλ,1(x),uλ,2(x)):=(uλ,vλ) is a classical solution of (1.1). By a classical uniqueness result (see, for instance, [7,Theorem 4.7]), (uλ,1,uλ,2) is a unique viscosity solution of (1.1). The rest of the claims in Corollary 2 is an immediate consequence of Theorem 1.

    Some remarks are in order. (ⅰ) Following [11], we may use Theorem 6 as the primary cornerstone for building a scalar Hamilton-Jacobi equation, for which the vanishing discount problem fails to have the full convergence as the discount factor goes to zero.

    (ⅱ) In the construction of the functions f,gC(R,R) in Theorem 6, the author has chosen d to satisfy d1, but, in fact, one may choose any d>0. In the proof, the core step is to find the function h(x)=f(x)g(x), with the properties: (a) the function xh(x)εx is nondecreasing on R for some ε>0 and (b) the curve y=h(x), with x<d, meets the lines y=p(xd) and y=q(xd), respectively, at Pj and Qj for all jN, where p,q,d are positive constants such that ε<p<q, and the sequences {Pj}jN,{Qj}jN converge to the point (d,0). Obviously, such a function h is never left-differentiable at x=d nor convex in any neighborhood of x=d. Because of this, it seems difficult to select f,gC(R,R) in Theorem 1, both smooth everywhere. In the proof of Theorem 6, we have chosen ε=k0, p=k, q=k2, Pj=(uνj,k(uνjd)), and Qj=(uμj,k2(uμjd))

    Another possible choice of h among many other ways is the following. Define first η:RR by η(x)=x(sin(log|x|)+2) if x0, and η(0)=0 (see Figure 2). Fix d>0 and set h(x)=η(xd) for xR. we remark that ηC(R{0}) and hC(R{d}). Note that if x0,

    η(x)=sin(log|x|)+cos(log|x|)+2[22,2+2],
    Figure 2.  Graph of η (slightly deformed).

    and that if we set xj=exp(2πj) and ξj=exp(2πj+π2), jN, then

    η(xj)=2xj   and   η(ξj)=3ξj.

    The points Pj:=(xj+d,2xj) are on the intersection of two curves y=h(x) and y=2(xd), while the points Qj:=(d+ξj,3ξj) are on the intersection of y=h(x) and y=3(xd). Moreover, limPj=limQj=(d,0).

    The author would like to thank the anonymous referees for their careful reading and useful suggestions. He was supported in part by the JSPS Grants KAKENHI No. 16H03948, No. 20K03688, No. 20H01817, and No. 21H00717.

    The author declares no conflict of interest.



    [1] Donoho DL (1995) De-noising by soft-thresholding. IEEE T Inform Theory 41: 613-627. doi: 10.1109/18.382009
    [2] Yao J, Zhang YT (2001) Bionic wavelet transform: A new time-frequency method based on an auditory model. IEEE T Biomed Eng 48: 856-863. doi: 10.1109/10.936362
    [3] Yao J, Zhang YT (2002) The application of bionic wavelet transforms to speech signal processing in cochlear implants using neural network simulations. IEEE T Biomed Eng 49: 1299-1309. doi: 10.1109/TBME.2002.804590
    [4] Yuan XL (2003) Auditory Model-based Bionic Wavelet Transform for Speech Enhancement. A thesis submitted to the graduate school in partial fulfillment.
    [5] Gold T (1948) Hearing II: The physical basis of the action of the cochlea. Proc Roy Soc B-Biol Sci 135: 492-498. doi: 10.1098/rspb.1948.0025
    [6] Xi Y, Bing-wu L, Fang Y (2010) Speech enhancement using bionic wavelet transform and adaptive threshold function. Second International Conference on Computational Intelligence and Natural Computing 1: 265-268.
    [7] Cohen MX (2019) A better way to define and describe Morlet wavelets for time-frequency analysis. Neuroimage 199: 81-86. doi: 10.1016/j.neuroimage.2019.05.048
    [8] Valencia D, Orejuela D, Salazar J, et al. (2016) Comparison analysis between rigrsure, sqtwolog, heursure and minimaxi techniques using hard and soft thresholding methods. XXI Symposium on Signal Processing, Images and Artificial Vision (STSIVA), 1-5.
    [9] Garg A, Sahu OP (2019) A hybrid approach for speech enhancement using Bionic wavelet transform and Butterworth filter. International Journal of Computers and Applications, 1-11.
    [10] Chen F, Zhang YT (2006) A new implementation of discrete bionic wavelet transform: Adaptive tiling. Digit Signal Process 16: 233-246. doi: 10.1016/j.dsp.2005.05.002
    [11] Mourad T, Lotfi S, Sabeur A, et al. (2009) Speech Enhancement with Bionic Wavelet Transform and Recurrent Neural Network. 5th International Conference: Sciences of Electronic, Technologies of Information and Telecommunications SETIT, 22-26.
    [12] Bin-Fang C, Jian-Qi L, Peixin Q, et al. (2014) An Optimization Adaptive BWT Speech Enhancement Method. Information Technology Journal 13: 1730-1736. doi: 10.3923/itj.2014.1730.1736
    [13] Mourad T (2016) Speech enhancement based on stationary bionic wavelet transform and maximum a posterior estimator of magnitude-squared spectrum. International Journal of Speech Technology 20: 75-88.
    [14] Wu LM, Li YF, Li FJ, et al. (2014) Speech Enhancement Based on Bionic Wavelet Transform and Correlation Denoising. Advanced Materials Research, 1386-1390.
    [15] Ben-Nasr M, Talbi M, Cherif A (2012) Arabic Speech Recognition by MFCC and Bionic Wavelet Transform using a Multi-Layer Perceptron for Voice Control. International Journal of Software Engineering and Technology.
    [16] Zehtabian A, Hassanpour H, Zehtabian S, et al. (2010) A novel speech enhancement approach based on Singular Value Decomposition and Genetic Algorithm. 2010 International Conference of Soft Computing and Pattern Recognition.
    [17] LIU Y, NI W (2015) Speech enhancement based on bionic wavelet transform of subband spectrum entropy. Journal of Computer Applications 3: 58.
    [18] Singh RA, Pritamdas K (2015) Enhancement of Speech Signal by Transform Method and Various threshold Techniques: A Literature Review. Advanced Research in Electrical and Electronic Engineering 2: 5-10.
    [19] Tan BT, Fu M, Spray A, et al. (1996) The use of wavelet Transforms in Phoneme Recognition. Proceeding of Fourth International Conference on Spoken Language Processing. ICSLP'96 4: 2431-2434. doi: 10.1109/ICSLP.1996.607300
    [20] Rioul O, Vetterli M (1991) Wavelets and Signal Processing. IEEE SP Magazine.
    [21] Kobayashi M and Sakamoto M (1993) Wavelets analysis of acoustic signals. Japan SIAM Wavelet Seminars II.
    [22] Jones DL, Baraniuk RG (1991) Efficient approximation of continuous wavelet transform. Electron Lett 27: 748-750. doi: 10.1049/el:19910465
    [23] Swami PD, Sharma R, Jain A, et al. (2015) Speech enhancement by noise driven adaptation of perceptual scales and thresholds of continuous wavelet transform coefficients. Speech Communication 70: 1-12. doi: 10.1016/j.specom.2015.02.007
    [24] Johnson MT, Yuan X, Ren Y (2007) Speech signal enhancement through adaptive wavelet thresholding. Speech Communication 49: 123-133. doi: 10.1016/j.specom.2006.12.002
    [25] Wavelet center frequency, MATLAB centfrq, MathWorks. Available from: https://in.mathworks.com/help/wavelet/ref/centfrq.html.
    [26] Anusuya MA and Katti SK (2011) Front end Analysis of Speech signal processing: A Review. International Journal of speech technology, Springer.
    [27] GitHub. Available from: https://github.com/Jakobovski/decoupled-multimodal-learning.
    [28] Hu Y and Loizou P (2007) Subjective evaluation and comparison of speech enhancement algorithms. Speech Communication 49: 588-601. doi: 10.1016/j.specom.2006.12.006
    [29] Aida-zade K, Xocayev A, Rustamov S (2016) Speech recognition using Support Vector Machines. 2016 IEEE 10th International Conference on Application of Information and Communication Technologies (AICT), 1-4.
    [30] Mini PP, Thomas T, Gopikakumari R (2018) Feature Vector Selection of Fusion of MFCC and SMRT Coefficients for SVM Classifier Based Speech Recognition System. 2018 8th International Symposium on Embedded Computing and System Design, 153-157.
    [31] Thiruvengatanadhan R (2018) Speech Recognition using SVM. International Research Journal of Engineering and Technology (IRJET) 5: 918-921.
    [32] Zou YX, Zheng WQ, Shi W, et al. (2014) Improved Voice Activity Detection based on support vector machine with high separable speech feature vectors. 2014 19th International Conference on Digital Signal Processing.
    [33] Gupta A, Joshi A (2018) Speech Recognition using Artificial Neural Network. International Conference on Communication and Signal Processing, India.
    [34] Al-Rababah MAA, Al-Marghilani A, Hamarshi AA (2018) Automatic Detection Technique for Speech Recognition based on Neural Networks Inter-Disciplinary. (IJACSA) International Journal of Advanced Computer Science and Applications 9: 179-184.
    [35] Swedia ER, Mutiara AB, Subali M (2018 ) Deep Learning Long-Short Term Memory (LSTM) for Indonesian Speech Digit Recognition using LPC and MFCC Feature. Third International Conference on Informatics and Computing (ICIC).
    [36] Vani HY, Anusuya MA (2015) Isolated Speech recognition using K-means and FCM Technique. International Conference on Emerging Research in Electronics, Computer Science and Technology (ICERECT).
    [37] Vani HY, Anusuya MA (2017) Noisy speech recognition using KFCM. International Conference on Cognitive Computing Information Processing.
    [38] ICSI Speech FAQ. Available from: https://www1.icsi.berkeley.edu/Speech/faq/speechSNR.html.
    [39] Yangyang S, Mei-Yuh H, Lei X (2019) END-TO-END SPEECH RECOGNITION USING A HIGH RANK LSTM-CTC BASED MODEL. ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).
    [40] Graves A, Jaitly N (2014) Towards End-to-End Speech Recognition with Recurrent Neural Networks. Proceedings of the 31 st International Conference on Machine Learning, Beijing, China.
    [41] Goehring T, Keshavarzi M, Carlyon RP, et al. (2019) Recurrent neural networks to improve the perception of speech in non-stationary noise by people with cochlear implants. J Acoust Soc Am 146: 705-718. doi: 10.1121/1.5119226
  • This article has been cited by:

    1. Yuandong Chen, Jinhao Pang, Yuchen Gou, Zhiming Lin, Shaofeng Zheng, Dewang Chen, Research on the A* Algorithm for Automatic Guided Vehicles in Large-Scale Maps, 2024, 14, 2076-3417, 10097, 10.3390/app142210097
  • Reader Comments
  • © 2020 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(5001) PDF downloads(568) Cited by(6)

Figures and Tables

Figures(4)  /  Tables(8)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog