#### DMCA

## Non-Causal Time-Domain Filters for Single-Channel Noise Reduction

Citations: | 2 - 1 self |

### Citations

778 |
Suppression of acoustic noise in speech using spectral subtraction
- Boll
- 1979
(Show Context)
Citation Context ...ise reduction methods, see, e.g., [1], [2], and the references therein. In general, we can divide these speech related noise reduction methods into three groups, i.e., spectral-subtractive algorithms =-=[3]-=-, statistical-model-based algorithms [4]–[7], and subspace algorithms [8]–[11]. The references, [3]–[5], and [8]–[10], refer to some of the pioneering work within these groups. Note that in the litera... |

688 |
Speech enhancement using a minimum-mean square error short-time spectral amplitude estimator
- Ephraim, Malah
- 1984
(Show Context)
Citation Context ... speech related noise reduction methods into three groups, i.e., spectral-subtractive algorithms [3], statistical-model-based algorithms [4]–[7], and subspace algorithms [8]–[11]. The references, [3]–=-=[5]-=-, and [8]–[10], refer to some of the pioneering work within these groups. Note that in the literature, noise reduction in speech applications is also termed speech enhancement. Often, noise reduction ... |

462 |
High-resolution frequency-wavenumber spectrum analysis
- Capon
- 1969
(Show Context)
Citation Context ...DUCTION 1531 C. Minimum Variance Distortionless Response The minimum variance distortionless response (MVDR) filter (a.k.a. the Capon filter) was proposed by Capon in the context of spatial filtering =-=[31]-=-, [32]. Here, the MVDR filter, or orthogonal decomposition-based MVDR (ODMVDR) filter as we term it, is used for temporal filtering, and it is designed on basis of the orthogonal decomposition. The OD... |

275 | Noise power spectral density estimation based on optimal smoothing and minimum statistics
- Martin
- 2001
(Show Context)
Citation Context ...ent transform domains. For example, it is possible to reduce the computational complexity by utilizing transform domain filters [12]. Two examples of transform domains are the Fourier [3], [9], [13], =-=[14]-=- and Karhunen–Loève [15], [16] domains. The filters can, though, be equivalently derived in all domains. In this paper, we consider time-domain filters only. Moreover, we restrict ourselves to the stu... |

275 | a fundamental frequency estimator for speech and music - Cheveigné, Kawahara, et al. - 2002 |

252 |
An algorithm for linearly constrained adaptive array processing,
- Frost
- 1972
(Show Context)
Citation Context ...last filter, described in this section, is the HDLCMV filter. This filter design is inspired by the LCMV beamformer (a.k.a. the Frost beamformer) proposed by Frost in the context of spatial filtering =-=[33]-=-. Here, we derive a non-causal HDLCMV filter for temporal filtering. The HDLCMV filter is designed to extract periodic signals modeled by (2), i.e., it is suited for extraction of signals such as voic... |

195 |
Trees, “A signal subspace approach for speech enhancement
- Ephraim, Van
- 1995
(Show Context)
Citation Context ...r in different transform domains. For example, it is possible to reduce the computational complexity by utilizing transform domain filters [12]. Two examples of transform domains are the Fourier [3], =-=[9]-=-, [13], [14] and Karhunen–Loève [15], [16] domains. The filters can, though, be equivalently derived in all domains. In this paper, we consider time-domain filters only. Moreover, we restrict ourselve... |

192 |
Speech enhancement using a soft-decision noise suppression filter
- Mcaulay, Malpass
- 1980
(Show Context)
Citation Context ...2], and the references therein. In general, we can divide these speech related noise reduction methods into three groups, i.e., spectral-subtractive algorithms [3], statistical-model-based algorithms =-=[4]-=-–[7], and subspace algorithms [8]–[11]. The references, [3]–[5], and [8]–[10], refer to some of the pioneering work within these groups. Note that in the literature, noise reduction in speech applicat... |

85 | Model-order selection: a review of information criterion rules - Stoica, Selen - 2004 |

76 |
Matrix theory.
- Franklin
- 1968
(Show Context)
Citation Context ...r an orthogonal decomposition based filter as (38) where (39) is the covariance matrix of the interference-plus-noise. The expression in (38) can also be recognized as a generalized Rayleigh quotient =-=[30]-=-. This quotient is maximized when the filter, , equals the eigenvector corresponding to the largest eigenvalue, , of . Clearly, is rank one, i.e., (40) with denoting the trace operator. An important o... |

61 |
Spetral analysis of signals. Upper Saddle River,
- Stoica, Moses
- 2005
(Show Context)
Citation Context ...s the covariance matrix of , and 1528 IEEE TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING, VOL. 20, NO. 5, JULY 2012 is the covariance matrix of . When is quasi-periodic, we can also model as =-=[29]-=- (7) with denoting the complex conjugate transpose operator, and (8) (9) (10) (11) (12) where denotes the construction of a diagonal matrix from a vector. The objective in traditional noise reduction ... |

54 |
Speech Enhancement, Theory and Practice. Boca
- Loizou
- 2007
(Show Context)
Citation Context ...of the noise in various signal processing applications. Over the years, numerous noise reduction methods have been proposed. For an overview of speech related noise reduction methods, see, e.g., [1], =-=[2]-=-, and the references therein. In general, we can divide these speech related noise reduction methods into three groups, i.e., spectral-subtractive algorithms [3], statistical-model-based algorithms [4... |

44 |
Speech enhancement from noise: A regenerative approach
- Dendrinos, Bakamidis, et al.
- 1991
(Show Context)
Citation Context ...n general, we can divide these speech related noise reduction methods into three groups, i.e., spectral-subtractive algorithms [3], statistical-model-based algorithms [4]–[7], and subspace algorithms =-=[8]-=-–[11]. The references, [3]–[5], and [8]–[10], refer to some of the pioneering work within these groups. Note that in the literature, noise reduction in speech applications is also termed speech enhanc... |

40 | Bayesian harmonic models for musical pitch estimation and analysis - Godsill, Davy |

36 | Subjective comparison and evaluation of speech enhancement algorithms
- Hu, Loizou
- 2007
(Show Context)
Citation Context ...ucted another real-life experiment on speech where we considered enhancement of female (sp12.wav) and male (sp02.wav) speech signals in noise. The speech signals are part of the NOIZEUS speech corpus =-=[37]-=-. Note that since the utilized speech signals contain segments of unvoiced speech, we only evaluate the ODW and ODMVDR filters in this experiment. Using the ODW and ODMVDR non-causal filtering schemes... |

23 | Computationally efficient parameter estimation for harmonic sinusoidal signals
- Li, Stoica, et al.
- 2000
(Show Context)
Citation Context ...armonic, and is the phase of the th harmonic. In this paper, we consider the pitch, , and the model order, , as known parameters. Numerous methods for estimation of these parameters exist [17], [18], =-=[22]-=-–[28]. Using Euler’s formula, we can also write (2) as (3) where is the complex amplitude of the th harmonic, and denotes the elementwise complex conjugate of a matrix/vector. To make the notation sim... |

21 |
Joint highresolution fundamental frequency and order estimation
- Christensen, Jakobsson, et al.
- 2007
(Show Context)
Citation Context ...ic, and is the phase of the th harmonic. In this paper, we consider the pitch, , and the model order, , as known parameters. Numerous methods for estimation of these parameters exist [17], [18], [22]–=-=[28]-=-. Using Euler’s formula, we can also write (2) as (3) where is the complex amplitude of the th harmonic, and denotes the elementwise complex conjugate of a matrix/vector. To make the notation simpler ... |

18 | Accurate frequency estimation for real harmonic sinusoids - Chan, So - 2004 |

18 |
Maximum-likelihood spectral estimation,”
- Capon
- 1983
(Show Context)
Citation Context ...N 1531 C. Minimum Variance Distortionless Response The minimum variance distortionless response (MVDR) filter (a.k.a. the Capon filter) was proposed by Capon in the context of spatial filtering [31], =-=[32]-=-. Here, the MVDR filter, or orthogonal decomposition-based MVDR (ODMVDR) filter as we term it, is used for temporal filtering, and it is designed on basis of the orthogonal decomposition. The ODMVDR f... |

16 |
Harmonic tunnelling: tracking non-stationary noises during speech,
- Ealey, Kelleher, et al.
- 2001
(Show Context)
Citation Context ...he noise statistics during silences by using a voice activity detector (VAD) if the noise is stationary, or we can estimate the noise statistics even in periods with voice activity using, e.g., [14], =-=[35]-=-. The ODW and ODMVDR filters were calculated using and , whereas the HDLCMV filter Fig. 7. Estimated log-spectral distances over time for causal and non-causal (a) ODW, (b) ODMVDR, and (c) HDLCMV filt... |

12 | Codebook-based Bayesian speech enhancement,”
- Srinivasan, Samuelsson, et al.
- 2005
(Show Context)
Citation Context ...and the references therein. In general, we can divide these speech related noise reduction methods into three groups, i.e., spectral-subtractive algorithms [3], statistical-model-based algorithms [4]–=-=[7]-=-, and subspace algorithms [8]–[11]. The references, [3]–[5], and [8]–[10], refer to some of the pioneering work within these groups. Note that in the literature, noise reduction in speech applications... |

12 |
Sørensen, Reduction of broad-band noise in speech by truncated QSVD
- Jensen, Hansen, et al.
- 1995
(Show Context)
Citation Context ...ed noise reduction methods into three groups, i.e., spectral-subtractive algorithms [3], statistical-model-based algorithms [4]–[7], and subspace algorithms [8]–[11]. The references, [3]–[5], and [8]–=-=[10]-=-, refer to some of the pioneering work within these groups. Note that in the literature, noise reduction in speech applications is also termed speech enhancement. Often, noise reduction methods rely o... |

10 |
Optimal Time-Domain Noise Reduction Filters – A Theoretical
- Benesty, Chen
(Show Context)
Citation Context ...eech. The filter can be derived directly in the time domain or in different transform domains. For example, it is possible to reduce the computational complexity by utilizing transform domain filters =-=[12]-=-. Two examples of transform domains are the Fourier [3], [9], [13], [14] and Karhunen–Loève [15], [16] domains. The filters can, though, be equivalently derived in all domains. In this paper, we consi... |

9 | A parametric method for pitch estimation of piano tones - Emiya, David, et al. - 2007 |

7 | Enhancement of single-channel periodic signals in the time-domain,”
- Jensen, Benesty, et al.
- 2012
(Show Context)
Citation Context ... 1558-7916/$31.00 © 2012 IEEE JENSEN et al.: NON-CAUSAL TIME-DOMAIN FILTERS FOR SINGLE-CHANNEL NOISE REDUCTION 1527 two decomposition approaches both have advantages and disadvantages as discussed in =-=[19]-=-. For example, the orthogonal decomposition based filters can be used for enhancing any kind of desired signal, however, they are sensible to nonstationary noise since it is difficult to estimate the ... |

6 |
Sparse Adaptive Filters for Echo Cancellation
- Paleologu, Benesty, et al.
- 2010
(Show Context)
Citation Context ... MVDR filters using the recursive procedure in Table I, we can reduce the computational complexity significantly compared to when the filters are calculated directly using (43) and (50), respectively =-=[34]-=-. Similar recursive JENSEN et al.: NON-CAUSAL TIME-DOMAIN FILTERS FOR SINGLE-CHANNEL NOISE REDUCTION 1533 filter expressions can be found for the non-causal filters where is between 0 and . Moreover, ... |

4 | Subspace-Based Noise Reduction for Speech Signals via Diagonal and Triangular
- Hansen, Jensen
- 1155
(Show Context)
Citation Context ...neral, we can divide these speech related noise reduction methods into three groups, i.e., spectral-subtractive algorithms [3], statistical-model-based algorithms [4]–[7], and subspace algorithms [8]–=-=[11]-=-. The references, [3]–[5], and [8]–[10], refer to some of the pioneering work within these groups. Note that in the literature, noise reduction in speech applications is also termed speech enhancement... |

4 | Study of the noise-reduction problem in the Karhunen-Loève expansion domain
- Chen, Benesty, et al.
- 2009
(Show Context)
Citation Context ...or example, it is possible to reduce the computational complexity by utilizing transform domain filters [12]. Two examples of transform domains are the Fourier [3], [9], [13], [14] and Karhunen–Loève =-=[15]-=-, [16] domains. The filters can, though, be equivalently derived in all domains. In this paper, we consider time-domain filters only. Moreover, we restrict ourselves to the study of single-channel fil... |

4 |
Classification, Parameter Estimation and State Estimation—An Engineering Approach using MATLAB
- Heijden, Duin, et al.
- 2004
(Show Context)
Citation Context ...er hand, was chosen on basis of an inspection of the spectrogram in Fig. 5(a). Furthermore, in the calculations of the HDLCMV filter, we regularized the covariance matrix of the observed signal as in =-=[36]-=- (93) The regularization is necessary due to estimation errors on the signal statistics and mismatch between the assumed harmonic model and the speech signal. We experienced that gives consistently go... |

2 | Rayleigh mixture model-based hidden markov modeling and estimation of noise in noisy speech signals - Sørensen, Andersen - 2007 |

2 |
Speech enhancement in the Karhunen-Loève expansion domain
- Benesty, Chen, et al.
- 2011
(Show Context)
Citation Context ...mple, it is possible to reduce the computational complexity by utilizing transform domain filters [12]. Two examples of transform domains are the Fourier [3], [9], [13], [14] and Karhunen–Loève [15], =-=[16]-=- domains. The filters can, though, be equivalently derived in all domains. In this paper, we consider time-domain filters only. Moreover, we restrict ourselves to the study of single-channel filters o... |

2 |
On singlechannel noise reduction in the time domain
- Chen, Benesty, et al.
(Show Context)
Citation Context ...ited for evaluation of non-causal filters. Moreover, we derive different non-causal orthogonal and harmonic decomposition based filters. Note that the causal filters proposed in [12], [17], [18], and =-=[21]-=- can be seen as special cases of the proposed designs. For the two particular cases where the filter is causal and anti-causal, respectively, we derive expressions for recursive updates of the orthogo... |

1 |
Multi-pitch estimation,” Synth
- Christensen, Jakobsson
- 2009
(Show Context)
Citation Context ...ltering schemes. was calculated using , the pitch estimated at every time instance, and a fixed harmonic model order of . We estimated the pitch using the orthogonality based subspace method in [17], =-=[18]-=- which is freely available online.1 The model order, on the other hand, was chosen on basis of an inspection of the spectrogram in Fig. 5(a). Furthermore, in the calculations of the HDLCMV filter, we ... |

1 |
Joint filtering scheme for nonstationary noise reduction,” 2012, unpublished manuscript
- Jensen, Benesty, et al.
(Show Context)
Citation Context ...same asymptotic performances for all s both with respect to noise reduction and distortion. This motivates using the orthogonal and harmonic decomposition based filters jointly as considered in [19], =-=[20]-=- for since they have complementary advantages and disadvantages. However, this will not be treated in this paper. VIII. EXAMPLE: NOISE REDUCTION OF SPEECH In the following, we demonstrate the applicab... |