Robust Feature Extraction for Continuous Speech Recognition Using the MVDR Spectrum Estimation Method

Dharanipragada, S.; Yapanel, Ümit; Rao, Bhaskar D.

doi:10.1109/tasl.2006.876776

Cited by 47 publications

(44 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…17 The class mean m c and global mean m were computed as: The within-class scatter matrices S W and the between-class scatter matrices S B were computed as:…”

Section: Class Separability Analysismentioning

confidence: 99%

See 1 more Smart Citation

Dynamic Feature Extraction for Speech Signal Based On MUSIC and Modulation Spectrum

Zhao¹,

Wang²

2016

IJAV

View full text Add to dashboard Cite

A novel dynamic feature extraction algorithm is proposed to help improve speech recognition robustness in noise environmental conditions. Owing to the modulation spectrum having time-frequency agglomeration performance, according to different reflections in the modulation spectrum for interference and speech signals, we first calculate the Multiple Signal Classification (MUSIC) spectrum and then get the modulation spectrum. We then filter the modulation spectrum signal. For the filter signal, use a 32 frames signal is used as a processing unit to get the modulation spectrum energy vector. This reveals the close correlation between the speech signal frames and can well reflect speech dynamic characteristics. Finally, the cepstrum coefficients are extracted as the feature parameter. It not only adequately reflects speech dynamic characteristics, but also has lower sensitivity for the speech environment. The effectiveness of the feature is discussed in view of the class separability and speaker variability properties. We evaluated the feature under different kinds of noise (white noise, pink noise, street noise, and panzer noise) and different signal-to-noise ratios (-5 dB, 0 dB, 5 dB, 10 dB, and 15 dB). The experimental results show that the novel feature has good robustness and computational efficiency under low signal-to-noise ratios and plays a very good foreshadowing role in latter speech research.

show abstract

“…17 The class mean m c and global mean m were computed as: The within-class scatter matrices S W and the between-class scatter matrices S B were computed as:…”

Section: Class Separability Analysismentioning

confidence: 99%

“…17 The following equation was used to define the mean of class c for speaker s: For the large databases, all the speakers had samples from all classes. Hence, M c will equal to S. We can now compute the class means as:…”

Section: Interspeaker Variability Analysismentioning

confidence: 99%

Dynamic Feature Extraction for Speech Signal Based On MUSIC and Modulation Spectrum

Zhao¹,

Wang²

2016

IJAV

View full text Add to dashboard Cite

show abstract

“…In CMS, the averages of all MFCC components are calculated, and then these averages are subtracted from MFCC [13] components. CMS can remove the channel efCopyright c 2012 The Institute of Electronics, Information and Communication Engineers fects happened in the convolutional distortion.…”

Section: Cmsmentioning

confidence: 99%

A Noise-Robust Continuous Speech Recognition System Using Block-Based Dynamic Range Adjustment

Sun

Miyanaga

2012

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

SUMMARYA new approach to speech feature estimation under noise circumstances is proposed in this paper. It is used in noise-robust continuous speech recognition (CSR). As the noise robust techniques in isolated word speech recognition, the running spectrum analysis (RSA), the running spectrum filtering (RSF) and the dynamic range adjustment (DRA) methods have been developed. Among them, only RSA has been applied to a CSR system. This paper proposes an extended DRA for a noise-robust CSR system. In the stage of speech recognition, a continuous speech waveform is automatically assigned to a block defined by a short time length. The extended DRA is applied to these estimated blocks. The average recognition rate of the proposed method has been improved under several different noise conditions. As a result, the recognition rates are improved up to 15% in various noises with 10 dB SNR.

show abstract

“…A detailed discussion of speech spectral estimation by MVDR can be found in [10], with focus on speech recognition and warped MVDR in [6], and with focus on robust feature extraction for recognition in [14].…”

Section: Spectralmentioning

confidence: 99%

Signal adaptive spectral envelope estimation for robust speech recognition

Wölfel

2009

Speech Communication

View full text Add to dashboard Cite

This paper describes a novel spectral envelope estimation technique which adapts to the characteristics of the observed signal. This is possible via the introduction of a second bilinear transformation into warped minimum variance distortionless response (MVDR) spectral envelope estimation. As opposed to the first bilinear transformation, however, which is applied in the time domain, the second bilinear transformation must be applied in the frequency domain. This extension enables the resolution of the spectral envelope estimate to be steered to lower or higher frequencies, while keeping the overall resolution of the estimate and the frequency axis fixed. When embedded in the feature extraction process of an automatic speech recognition system, it provides for the emphasis of the characteristics of speech features that are relevant for robust classification, while simultaneously suppressing characteristics that are irrelevant for classification. The change in resolution may be steered, for each observation window, by the normalized first autocorrelation coefficient.To evaluate the proposed adaptive spectral envelope technique, dubbed warped-twice MVDR, we use two objective functions: class separability and word error rate. Our test set consists of development and evaluation data as provided by NIST for the Rich Transcription 2005 Spring Meeting Recognition Evaluation. For both measures, we observed consistent improvements for several speaker-to-microphone distances. In average, over all distances, the proposed front-end reduces the word error rate by 4% relative compared to the widely used mel-frequency cepstral coefficients as well as perceptual linear prediction.

show abstract

Robust Feature Extraction for Continuous Speech Recognition Using the MVDR Spectrum Estimation Method

Cited by 47 publications

References 23 publications

Dynamic Feature Extraction for Speech Signal Based On MUSIC and Modulation Spectrum

Dynamic Feature Extraction for Speech Signal Based On MUSIC and Modulation Spectrum

A Noise-Robust Continuous Speech Recognition System Using Block-Based Dynamic Range Adjustment

Signal adaptive spectral envelope estimation for robust speech recognition

Contact Info

Product

Resources

About