Robust speech recognition based on stochastic matching

Sankar, Ananth; Lee, Chin‐Hui

doi:10.1109/icassp.1995.479288

Cited by 47 publications

(21 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Stochastic Matching algorithm (Sankar and Lee, 1995) is the closest in spirit to the STAR algorithm when no stereo data are provided. Some of our solutions are identical to the ones provided by the stochastic matching algorithm.…”

Section: Comparison With Previous Algorithmsmentioning

confidence: 99%

“…When particularized for data compensation these solutions can be directly compared to ®xed codeword-dependent cepstral normalization (FCDCN) (Acero, 1991). When particularized for HMM adaptation they can be compared to dual-channel codebook adaptation (DCCA) , maximum likelihood linear regression (MLLR) (Leggeter and Woodland, 1995), parallel model combination (PMC) (Gales, 1995), and stochastic matching (Sankar and Lee, 1995).…”

Section: Comparison With Previous Algorithmsmentioning

confidence: 99%

“…Acero and Stern, 1990;Liu et al, 1994;Neumeyer and Weintraub, 1994;Moreno et al, 1996) while other techniques have modi®ed parameters characterizing the statistical model of speech sounds such as the means and covariances of the distributions of HMMs (e.g. Gales, 1995;Sankar and Lee, 1995). However, both kind of approaches have usually been presented as separate and distinct techniques.…”

Section: Introductionmentioning

confidence: 99%

“…Many research groups have described algorithms that achieve a greater degree of environmental robustness using a variety of techniques. For example, Acero and Stern (1990), Varga and Moore (1990), Ephraim (1992), Gales and Young (1993), Leggeter and Woodland (1995), and Sankar and Lee (1995) have all described compensation strategies that involve the use of a para-metric model of degradation, combined with optimal estimation of the parameters of the model. Other approaches to robustness include the use of peripheral features based on models of human perception of speech signals (e.g.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Data-driven environmental compensation for speech recognition: A unified approach

Moreno¹,

Raj²,

Stern³

1998

Speech Communication

View full text Add to dashboard Cite

Environmental robustness for automatic speech recognition systems based on parameter modi®cation can be accomplished in two complementary ways. One approach is to modify the incoming features of environmentally-degraded speech to more closely resemble the features of the (normally undegraded) speech used to train the classi®er. The other approach is to modifying the internal statistical representations of speech features used by the classi®er to more closely resemble the features representing degraded speech in a particular target environment. This paper attempts to unify these two approaches to robust speech recognition by presenting several techniques that share the same basic assumptions and internal structure while diering in whether they modify the features of incoming speech or whether they modify the statistics of the classi®er itself. We present the multivaRiate gAussian-based cepsTral normaliZation (RATZ) family of algorithms which modify incoming cepstral features, along with the STAR (STAtistical Reestimation) family of algorithms, which modify the internal statistics of the classi®er. Both types of algorithms are data driven, in that they make use of a certain amount of adaptation data for learning compensation parameters. The algorithms were evaluated using the SPHINX-II speech recognition system on subsets of the Wall Street Journal database. While all algorithms demonstrated improved recognition accuracy compared to previous algorithms, the STAR family of algorithms tended to provide lower error rates than the RATZ family of algorithms as the SNR was decreased. Ó 1998 Elsevier Science B.V. All rights reserved. Re Âsume ÂDans les syst emes de reconnaissance de la parole, la robustesse a l'environnement par adaptation des param etres peutêtre obtenue de deux fac Ëons compl ementaires. Une premi ere approche consiste a modi®er les param etres acoustiques de la parole d egrad ee par l'environnement de fac Ëon a ressembler aux param etres de la parole (habituellement non d egrad ee) qui a et e utilis ee lors de l'entraõnement. La deuxi eme solution est de modi®er les param etres statistiques internes au reconnaisseur de fac Ëon a mieux repr esenter les caract eristiques de la parole d egrad ee dans un environnement cible particulier. Le pr esent papier tente d'uni®er ces deux approches de reconnaissance robuste de la parole en pr esentant plusieurs techniques qui partagent les mêmes hypoth eses de base et la même structure, tout en di erant dans le choix de savoir si elles modi®ent les param etres d'entr ee ou les param etres statistiques du reconnaisseur. Nous pr esentons ici la famille d'algorithmes bas es sur la normalisation cepstrale gaussienne multi-variable (RATZ) qui modi®ent les caract eristiques cepstrales d'entr ee, ainsi que les algorithmes STAR (re-estimation statistique), qui modi®ent les param etres internes du reconnaisseur. Les deux types d'algorithmes sont bas es sur les donn ees et utilisent une certaine quantit e de donn ee d'adaptation pour estimer les param etres de compen...

show abstract

Section: Comparison With Previous Algorithmsmentioning

confidence: 99%

Section: Comparison With Previous Algorithmsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Data-driven environmental compensation for speech recognition: A unified approach

Moreno¹,

Raj²,

Stern³

1998

Speech Communication

View full text Add to dashboard Cite

show abstract

“…Secondly there is the possibility of adapting from sparse data, for which ML is unable to provide robust estimations, originally solved by using prior information in the training process (MAP, maximum a posteriori) [Chesta et al, 1999]. More frameworks have been proposed such as Cepstral mean normalization [Liu et al, 1993], stochastic matching [Sankar and Lee, 1995], vector field smoothing [Tonomura et al, 1995], but they are beyond the purpose of this research.…”

Section: Adaptation Techniquesmentioning

confidence: 99%

Design and Evaluation of Statistical Parametric Techniques in Expressive Text-To-Speech: Emotion and Speaking Styles Transplantation

Trueba¹

View full text Add to dashboard Cite

Model adaptation by HMM decomposition and composition in noisy reverberant environments

2000

View full text Add to dashboard Cite

In hands‐free speech recognition in which the user speaks at a distance from the microphone, the accuracy of recognition is degraded if the environment is reverberant. The reason for the degradation is that the uttered speech is affected by the surrounding noise and reverberation, which produces a mismatch between the training data and the observed data. In order to cope with such a situation, the authors proposed a speech recognition method by HMM composition, involving the composition of an acoustic transfer function HMM [1, 2]. In that method, however, the acoustic transfer functions must be measured from various points before recognition. This paper proposes a method in which the acoustic transfer function HMM is estimated from the observed signal. In the proposed method, it is not required that the position of the speaker be known. Using adaptation data uttered from an arbitrary point, the HMM is decomposed into the known HMM and another HMM by maximum‐likelihood estimation, and the model parameters are estimated. As a result of a phonemewise 500‐word recognition experiment, the speaker‐dependent recognition rate is improved from 77.2% to 91.2%, and the speaker‐independent recognition rate is improved from 54.4% to 66.2%, demonstrating the effectiveness of the proposed method. © 2000 Scripta Technica, Syst Comp Jpn, 31(5): 77–85, 2000

show abstract

Robust speech recognition based on stochastic matching

Cited by 47 publications

References 11 publications

Data-driven environmental compensation for speech recognition: A unified approach

Data-driven environmental compensation for speech recognition: A unified approach

Design and Evaluation of Statistical Parametric Techniques in Expressive Text-To-Speech: Emotion and Speaking Styles Transplantation

Model adaptation by HMM decomposition and composition in noisy reverberant environments

Contact Info

Product

Resources

About