Normalizing the Speech Modulation Spectrum for Robust Speech Recognition

Xiao, Xiong; Chng, Eng Siong; Li, Haizhou

doi:10.1109/icassp.2007.367246

Cited by 14 publications

(8 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this case, there are only K free parameters. It is reported in [61] that the ensemble modeling approach outperforms MLLR significantly on Aurora-2 task. Two extensions of ensemble modeling are reported in [79].…”

Section: Ensemble Modelingmentioning

confidence: 99%

See 1 more Smart Citation

Features and Model Adaptation Techniques for Robust Speech Recognition: A Review

Legoh¹,

Bhattacharjee²,

Tuithung³

2015

CAE

View full text Add to dashboard Cite

In this paper, major speech features used in state-of-the-art technology in speech recognition research are reviewed. Also a brief review of major technological advancements during last few decades and a trend towards development of robust speech recognition system in terms of feature and model adaptation techniques is given. It has been the dream of researchers to develop a machine that recognizes speech and understands natural language like human but the reality is that the performance of the speech recognition system drastically degrades due to various adverse conditions like noise, variability in speaker, channel, device and mismatches in training and testing. This paper may be useful as a tutorial and review on state-of-the-art techniques for feature selection, feature normalization and model adaptation techniques for development of robust speech recognition system.

show abstract

Section: Ensemble Modelingmentioning

confidence: 99%

“…Hence, it is coarse to normalize the histogram of just one utterance to the global histogram of the entire training set which usually consists of thousands of utterances. There is another simple example to demonstrate the drawbacks of CMN, CVN, and HEQ [61]. Suppose there is an utterance with several words.…”

Section: Histogram Normalizationmentioning

confidence: 99%

Features and Model Adaptation Techniques for Robust Speech Recognition: A Review

Legoh¹,

Bhattacharjee²,

Tuithung³

2015

CAE

View full text Add to dashboard Cite

show abstract

“…The temporal filters are another group of techniques to reduce the mismatch by filtering the feature trajectories [4][5][6][7][8][9][10][11][12]. These temporal filters are design from different perspectives, e.g., the RelAtive SpecTrA (RASTA) filter [4] and the MVA filter [5] are empirically designed to remove the very low and/or high modulation frequencies that are believed to be less relevant to speech intelligibility but prone to environmental distortions.…”

Section: Introductionmentioning

confidence: 99%

“…The data-driven filters [6][7][8] are designed from the data to better represent the features or improve the features' discriminative ability. The temporal structure normalization (TSN) filter [9][10][11] and the temporal smoothing (TES) filter [12] are designed to normalize the inter-frame correlation of the feature. Although designed from very different criteria, the resulting temporal filters are usually band-pass or low-pass.…”

Section: Introductionmentioning

confidence: 99%

Effect of Feature Smoothing for Robust Speech Recognition

Xiao

Chng

2008

2008 6th International Symposium on Chinese Spoken Language Processing

Self Cite

View full text Add to dashboard Cite

One class of feature enhancement techniques improve features' robustness by performing temporal filtering to smooth the feature trajectories. While smoothing can enhance the features' robustness by reducing the intra-class variation of the features, it also compromises the features' discriminative power by reducing their inter-class distance. In this paper, we investigate the effect of feature smoothing on speech recognition performance. To evaluate how different degrees of smoothing will affect the performance, the speech features are low-pass filtered with different cut-off frequencies and then used for model training and recognition. From the experimental results, we have two observations: 1) the noisy speech needs more aggressive feature smoothing; 2) the large vocabulary Aurora-4 task prefers less smoothing than the small vocabulary Aurora-2 task.

show abstract

“…Compared to other temporal filters [27][28][29][30][31][32][33][34], an advantage of the TSN filter is that it is able to automatically adapt to environment distortions in the speech signal and provides customized filtering. The work to be presented in this chapter has been published in [128][129][130][131].…”

mentioning

confidence: 99%

Robust speech features and acoustic models for speech recognition

Xiao¹

Self Cite

View full text Add to dashboard Cite

This thesis examines techniques to improve the robustness of automatic speech recognition (ASR) systems against noise distortions. The study is important as the performance of ASR systems degrades dramatically in adverse environments, and hence greatly limits the speech recognition application deployment in realistic environments. Towards this end, we examine a feature compensation approach and a discriminative model training approach to improve the robustness of speech recognition system. The degradation of recognition performance is mainly due to the statistical mismatch between clean-trained acoustical model and noisy testing speech features. To reduce the feature-model mismatch, we propose to normalize the temporal structure of both training and testing speech features. Speech features' temporal structures are represented by the power spectral density (PSD) functions of feature trajectories. We propose to normalize the temporal structures by applying equalizing filters to the feature trajectories. The proposed filter is called temporal structure normalization (TSN) filter. Compared to other temporal filters used in speech recognition, the advantage of the TSN filter is its adaptability to changing environments. The TSN filter can also be viewed as a feature normalization technique that normalizes the PSD function of features, while other normalization methods, such as histogram equalization (HEQ), normalize the probability density function (p.d.f.) of features. Experimental study shows that the TSN filter produces better performance than other state-of-the-art temporal filters on both small vocabulary Aurora-2 task and large vocabulary Aurora-4 task. In the second study, we improve the robustness of speech recognition by improving the generalization capability of acoustic model rather than reducing the feature-model mismatch. In the log likelihood score domain, noise distortion causes the log likelihood score of noisy features to deviate from that of clean features. The deviation may move 6 ATTENTION: The Singapore Copyright Act applies to the use of this document. Nanyang Technological University Library the noisy features to the wrong side of the decision boundary that is trained from clean features, and hence causes recognition error. To improve performance, discriminative training (DT) methods, including minimum classification error (MCE), maximum mutual information (MMI) and soft-margin estimation (SME), are applied to improve the generalization capability of the acoustic model, which in turn is implemented by increasing the margin, i.e. the desired minimum distance from training samples to the decision boundary. Experimental study shows that by improving the acoustic model's generalization capability with SME and other DT training, speech recognition performance can be improved even when the testing data is mismatched from the training data. IN addition, the margin-based SME is slightly more effective than MCE and MMI in terms of increasing the margin and robustness. It is also observed that DT methods ...

show abstract

Normalizing the Speech Modulation Spectrum for Robust Speech Recognition

Cited by 14 publications

References 43 publications

Features and Model Adaptation Techniques for Robust Speech Recognition: A Review

Features and Model Adaptation Techniques for Robust Speech Recognition: A Review

Effect of Feature Smoothing for Robust Speech Recognition

Robust speech features and acoustic models for speech recognition

Contact Info

Product

Resources

About