Feature extraction based on DCT and MVDR spectral estimation for robust speech recognition

Seyedin, Sanaz; Ahadi, Mohammad

doi:10.1109/icosp.2008.4697205

Cited by 7 publications

(9 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Then the warped MVDR power spectrum is computed from the socalled Mel-warped spectrum similar to the way parameters are calculated in PMCC, after applying weighting to subbands. Then, the cepstral features are calculated by applying IFFT to the Mel-scale MVDR log-spectrum [6], [7]. The Mel-warped spectrum is also known as sub-band spectrum in the area of speech recognition.…”

Section: The Proposed New Front-end: Rpmcc Featuresmentioning

confidence: 99%

“…However, it has been shown that the MVDR method, by itself, is not as efficient as expected in low signal to noise ratios [6]. Therefore, an improvement in the feature extraction process based on MVDR is sought in this paper.…”

Section: Introductionmentioning

confidence: 99%

“…Robust speech recognition methods may be classified into four main categories [1]: speech signal to make it robust against additive or convolutional distortions is more widely used, since the speech features are mostly derived either directly from the power spectrum of the speech signal or from a modified version of it. Mel Frequency Cepstral Coefficients (MFCC) [2], Perceptual Linear Prediction (PLP) [3], Differential Power Spectrum (DPS) [4], Autocorrelation Mel Frequency Cepstral Coefficients (AMFCC) [5], DCT and MVDR features [6], [7] are good examples in this category. Furthermore, feature normalization techniques are considered as one of the most significant group of methods in the post-processing of features.…”

Section: Introductionmentioning

confidence: 99%

“…Bias is Copyright c 2010 The Institute of Electronics, Information and Communication Engineers mainly caused by the leakage of power from surrounding frequencies of the band-pass filter used to measure the power [7]. This problem can be solved by using DCT instead of FFT [6]. Moreover, large variance is due to employing a single sample in the power estimation process [7].…”

Section: Introductionmentioning

confidence: 99%

“…Moreover, large variance is due to employing a single sample in the power estimation process [7]. Both of these shortcomings have been addressed by the MVDR spectrum estimation method [6], [7]. During the use of stochastic models for acoustic modeling, such as hidden Markov models (HMMs), the state parameter distributions are usually modeled by mixtures of Gaussians.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

A New Subband-Weighted MVDR-Based Front-End for Robust Speech Recognition

Seyedin

Ahadi

2010

IEICE Trans. Inf. & Syst.

Self Cite

View full text Add to dashboard Cite

SUMMARYThis paper presents a novel noise-robust feature extraction method for speech recognition. It is based on making the Minimum Variance Distortionless Response (MVDR) power spectrum estimation method robust against noise. This robustness is obtained by modifying the distortionless constraint of the MVDR spectral estimation method via weighting the sub-band power spectrum values based on the sub-band signal to noise ratios. The optimum weighting is obtained by employing the experimental findings of psychoacoustics. According to our experiments, this technique is successful in modifying the power spectrum of speech signals and making it robust against noise. The above method, when evaluated on Aurora 2 task for recognition purposes, outperformed both the MFCC features as the baseline and the MVDR-based features in different noisy conditions.

show abstract

Section: The Proposed New Front-end: Rpmcc Featuresmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

A New Subband-Weighted MVDR-Based Front-End for Robust Speech Recognition

Seyedin

Ahadi

2010

IEICE Trans. Inf. & Syst.

Self Cite

View full text Add to dashboard Cite

show abstract

Audio-Visual Speech Processing for Human Computer Interaction

Chin

Seng

Ang

2012

Intelligent Systems Reference Library

View full text Add to dashboard Cite

Speech Enhancement Using Joint DNN-NMF Model Learned with Multi-Objective Frequency Differential Spectrum Loss Function

Pashaian,

Seyedin

2024

IET Signal Processing

Self Cite

View full text Add to dashboard Cite

We propose a multi-objective joint model of non-negative matrix factorization (NMF) and deep neural network (DNN) with a new loss function for speech enhancement. The proposed loss function (LMOFD) is a weighted combination of a frequency differential spectrum mean squared error (MSE)-based loss function (LFD) and a multi-objective MSE loss function LMO. The conventional MSE loss function computes the discrepancy between the estimated speech and clean speech across all frequencies, disregarding the process of changing amplitude in the frequency domain which contains valuable information. The differential spectrum representation retains spectral peaks that carry important information. Using this representation helps to ensure that this information in the speech signal is reserved. Also, on the other hand, noise spectra typically have a flat shape and as the differential operation makes the flat spectral partly close to zero, the differential spectrum is resistant to noises with smooth structures. Thus, we propose using a frequency-differentiated loss function that considers the magnitude spectrum differentiations between the neighboring frequency bins in each time frame. This approach maintains the spectrum variations of the objective signal in the frequency domain, which can effectively reduce the noise deterioration effects. The multi-objective MSE term LMO is a combined two-loss function related to the NMF coefficients which are the intermediate output targets, and the original spectral signals as the actual output targets. The use of encoded NMF coefficients as low-dimensional structural features for DNN serves as prior knowledge and helps the learning process. LMO is used beside LFD to take advantage of both the properties of the original and the differential spectrum in the training loss function. Moreover, a DNN-based noise classification and fusion strategy (NCF) is proposed to exploit a discriminative model for noise reduction. The experiments reveal the improvements of the proposed approach compared to the previous methods.

show abstract

Feature extraction based on DCT and MVDR spectral estimation for robust speech recognition

Cited by 7 publications

References 7 publications

A New Subband-Weighted MVDR-Based Front-End for Robust Speech Recognition

A New Subband-Weighted MVDR-Based Front-End for Robust Speech Recognition

Audio-Visual Speech Processing for Human Computer Interaction

Speech Enhancement Using Joint DNN-NMF Model Learned with Multi-Objective Frequency Differential Spectrum Loss Function

Contact Info

Product

Resources

About