Compensation for the effect of the communication channel in auditory-like analysis of speech (RASTA-PLP)

Heřmanský, Hynek; Morgan, Nelson; Bayya, Aruna; Kohn, Phil

doi:10.21437/eurospeech.1991-312

Cited by 98 publications

(11 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Various similar filtering techniques that attempt to remove slowly varying channel biases in the logarithmic-energy of cepstrum domain have been proposed and successfully applied to increase the noise robustness. Among others there are RelAtive SpecTrAl (RASTA) filtering [Hermansky et al 1991, Hermansky andMorgan 1994], phase-corrected RASTA [ de Veth and Boves 1996] and LDA-derived RASTA filtering [van Vuuren and Hermansky 1997], high-pass filtering [Hirsch et al 1991] and the Gaussian dynamic cepstrum representation [Aikawa et al 1993, Singer et al 1995. Combinations of different methods were investigated in [Junqua et al 1995].…”

Section: Filtering Techniquesmentioning

confidence: 99%

Quantile based histogram equalization for noise robust speech recognition

Hilger¹,

Ney²

2001

7th European Conference on Speech Communication and Technology (Eurospeech 2001)

View full text Add to dashboard Cite

This work describes an algorithm to increase the noise robustness of automatic speech recognition systems.In many practical applications recognition systems have to work in adverse acoustic environment conditions. Distortions and noises caused by the transmission are typical for telephone applications. Considerable amounts of varying background noise are a problem for all mobile applications such as cellular phones or speech controlled systems in cars. Automatic systems are much more sensitive to the variabilities of the acoustic signal than humans. Whenever there is a mismatch between the distribution of the training data and the data that is to be recognized, the recognition word error rates will increase. Verzögerung als bei einer konventionellen Mittelwerts-und Varianznormierung erforderlich ist. Um die Verallgemeinerbarkeit des Verfahrens zu untersuchen, wurden Experimente mit verschiedenen Spracherkennungssystemen auf unterschiedlichen Datensätzen durchgeführt: von Ziffernketten (SpeechDat Car) bis hin zu Erkennungsaufgaben mit einem größeren Vokabular von Einzelworten (Car Navigation) und kontinuierlicher Sprache (Wall Street Journal mit Geräuschen unterlegt). Auf allen Datensätzen wurden konsistente Erkennungsergebnisse beobachtet. Die modifizierte Merkmalsextraktion, mit der Wurzelfunktion an Stelle des Logarithmus, lieferte auf verrauschten Daten bereits bessere Erkennungsergebnisse als das Original. Die auf Quantilen basierende Transformation individueller Filterkanäle konnte diese Ergebnisse immer verbessern, abhängig von der Erkennungsaufgabe und dem Missverhältnis zwischen den Trainings-und Testdaten lagen die relativen Verbesserungen zwischen 5% und 50%. Schließlich konnte die Kombination benachbarter Filterkanäle die Fehlerrate noch etwas weiter reduzieren, insbesondere bei bandbegrenzten Geräuschen wie beispielsweise Fahrgeräuschen in Autos.

show abstract

Section: Filtering Techniquesmentioning

confidence: 99%

Quantile based histogram equalization for noise robust speech recognition

Hilger¹,

Ney²

2001

7th European Conference on Speech Communication and Technology (Eurospeech 2001)

View full text Add to dashboard Cite

show abstract

“…We applied three CN techniques to the twelve MFCC coordinates of the feature vector in this paper. We either used RASTA with integration factor 0.98 [2,3], or the GDCR approach [4] or CMS [1]. We kept the original values of delta-MFCC's, logE and delta-logE.…”

Section: Signal Processingmentioning

confidence: 99%

“…For automatic speech recognition over telephone lines it is wellknown that recognition performance can be seriously degraded due to the transfer characteristics of the communication channel. In order to reduce the influence of the linear filtering effect of the telephone handset and telephone line, different channel normalisation (CN) techniques have been proposed [for example 1,2,3,4]. Several studies addressed the question of the relative effectiveness of different CN approaches [for example 5,6].…”

Section: Introductionmentioning

confidence: 99%

“…We studied three different CN techniques in the context of a connected digit recognition task: cepstrum mean substraction (CMS) [1], RASTA filtering [2,3], and the Gaussian dynamic cepstrum representation (GDCR) [4]. For this task, we used hidden Markov models (HMMs) with Gaussian mixture densities describing the output probability density function of each state.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Comparison of channel normalisation techniques for automatic speech recognition over the phone

Veth¹,

Boves²

1996

4th International Conference on Spoken Language Processing (ICSLP 1996)

View full text Add to dashboard Cite

We compared three different channel normalisation (CN) methods in the context of a connected digit recognition task over the phone: ceptrum mean substraction (CMS), RASTA filtering and the Gaussian dynamic cepstrum reprsentation (GDCR). Using a small set of context-independent (CI) continuous Gaussian mixture hidden Markov models (HMMs) we found that CMS and RASTA outperformed the GDCR technique. We show that the main cause for the superiority of CMS compared to RASTA is the phase distortion introduced by the RASTA filter. Recognition results for a phasecorrected RASTA technique are identical to those of CMS. Our results indicate that an ideal cepstrum based CN method should (1) effectively remove the DC-component, (2) at least preserve modulation frequencies in the range 2-16 Hz and (3) introduce no phase distortion in case CI HMMs are used for recognition.

show abstract

“…Accordingly, the codeword-dependent cepstral normalization (CDCN) [2] and RASTA method [3] were presented for reducing the variability of additive noise and channel effect. Besides, the signal bias removal (SBR) [4], stochastic matching (SM) [5] and channel-effect-cancellation [6] methods were also successfully applied for telephone speech recognition.…”

Section: Introductionmentioning

confidence: 99%

Estimation of channel bias for telephone speech recognition

Chien¹,

Wang²,

Lee³

1996

4th International Conference on Spoken Language Processing (ICSLP 1996)

View full text Add to dashboard Cite

In this study, we propose a maximum a posterior (MAP) estimation of channel bias to compensate the channel mismatch in telephone speech recognition. For a telephone speech, the channel bias is estimated by maximizing a posterior probability. Because a posterior probability is composed of a likelihood function and a prior density, we introduce a scale factor to evaluate their weights in MAP estimation. To further improve the performance, a prior channel statistics is extended to multiple components and the channel mismatch is separately compensated for different segments. Besides, a rapid MAP estimation applied in feature domain is also proposed for reducing the computational complexity. Experiments show that proposed method can significantly improve recognition rates and computational complexity.

show abstract

Compensation for the effect of the communication channel in auditory-like analysis of speech (RASTA-PLP)

Cited by 98 publications

References 0 publications

Quantile based histogram equalization for noise robust speech recognition

Quantile based histogram equalization for noise robust speech recognition

Comparison of channel normalisation techniques for automatic speech recognition over the phone

Estimation of channel bias for telephone speech recognition

Contact Info

Product

Resources

About