An integrated study of speaker normalisation and HMM adaptation for noise robust speaker-independent speech recognition

Hariharan, Ramalingam; Viikki, Olli

doi:10.1016/s0167-6393(01)00039-5

Cited by 3 publications

(4 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A transformation is used to minimize the mismatch between new utterances and average utterances. This study shows simple and commonly used speaker adaptation techniques (MLLR (Goronzy et al, 2004;Giuliani et al, 2006), CMLLR (Hariharan et al, 2002;Sundermann et al, 2003;Legetter and Woodland, 1995;Shen and Reynolds, 2008) and VTLN) to compensate for speaker-specific differences caused by non-native language influence for isolated words. Table 2 shows the results of the baseline system adapted with speaker adaptation techniques (MLLR, CMLLR and VTLN).…”

Section: Acoustic Model Adaptationmentioning

confidence: 99%

“…Perceptual-based evaluation of human raters is not only to simply value non-native utterances as accepted/rejected but also to analyze and locate specific errors on segmental aspects. Further, the acoustic model adaptation is combined with three speaker adaptation techniques Maximum Likelihood Linear Regression (MLLR) as proposed in (Goronzy et al, 2004;Giuliani et al, 2006;Haraty and El Ariss, 2007), Constrained MLLR (CMLLR) and Vocal Track Length ormalization (VTLN) as proposed in (Hariharan et al, 2002;Sundermann et al, 2003;Legetter and Woodland, 1995;Shen and Reynolds, 2008;Al-Haddad et al, 2009;Gales and Young, 2008) in order to eliminate interspeaker variability. Performance of the proposed acoustic model adaptation is evaluated in five measures of alignment analysis between recognition results and perceptual based evaluation: Hit, False Alarm (FA), Miss, Rejection and Hit + Rejection.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Acoustic Model Adaptation for Indonesian Language Utterance Training System

Linda¹,

Chisaki

Usagawa

2010

Journal of Computer Science

View full text Add to dashboard Cite

Problem statement: In order to build an utterance training system for Indonesian language, a speech recognition system designed for Indonesian is necessary. However, the system hardly works well due to the pronunciation variants of non-native utterances may lead to substitution/deletion error. This research investigated the pronunciation variant and proposes acoustic model adaptation to improve performance of the system. Approach: The proposed acoustic model adaptation worked in three steps: to analyze pronunciation variant with knowledge-based and data-derived methods; to align knowledge-based and data-derived results in order to list frequently mispronounced phones with their variants; to perform a state-clustering procedure with the list obtained from the second step. Further, three Speaker Adaptation (SA) techniques were used in combination with the acoustic model adaptation and they are compared each other. In order to evaluate and tune the adaptation techniques, perceptual-based evaluation by three human raters is performed to obtain the "true"recognition results. Results: The proposed method achieved an average gain in Hit + Rejection (the percentage of correctly accepted and correctly rejected utterances by the system as the human raters do) of 2.9 points and 2 points for native and non-native subjects, respectively, when compared with the system without adaptation. Average gains of 12.7 and 6.2 points for native and non-native students in Hit + Rejection were obtained by combining SA to the acoustic model adaptation. Conclusion/Recommendations: Performance evaluation of the adapted system demonstrated that the proposed acoustic model adaptation can improve Hit even though there is a slight increase of False Alarm (FA, the percentage of incorrectly accepted utterances by the system of which the human raters reject). The performance of the proposed acoustic model adaptation depends strongly on the effectiveness of state-clustering procedure to recover only in-vocabulary words. For future research, a confidence measure to discriminate between in-vocabulary and out-vocabulary words will be investigated

show abstract

Section: Acoustic Model Adaptationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Acoustic Model Adaptation for Indonesian Language Utterance Training System

Linda¹,

Chisaki

Usagawa

2010

Journal of Computer Science

View full text Add to dashboard Cite

show abstract

“…Especially when the handling of speech sound field, HMM has gotten thorough research and specific application. Now, it has gotten fairly good identification effect in isolating word identification, joins word identification and continuous speech recognition [2]. HMM has the very strong ability and sequential pattern classification of building mould for development process time sequence, can solve the random uncertain problem and suits the analysis of unsteady repetitive signal that has not good repeatability especially.…”

Section: Hmm Theorymentioning

confidence: 99%

Study of On-Line Condition Monitoring System for Roller Based on HMM

Tang

Lin

Zhang

2010

AMR

View full text Add to dashboard Cite

A kind of signal build mould and identification tool for roller of rolling mill was discussed. Basic algorithm of Hidden Markov model (HMM) theory was presented. For ψ500×2 shrewd roller of rolling mill, the paper led Short-time Fourier transform (STFT) into draw the feature of roller signal and chose Continues Gaussian Hidden Markov Model (CGHMM) to build and identify mould. The experiment installation major included: ψ500×2 shrewd rollers and roller performance test control work of rolling mill control machine as well as a set of online state inspection system. The experiment data was collected by two kinds: roller rotational speed from low to high and from high to low. The experiment shows that this system has better identification rate, and proves that this system is effective.

show abstract

“…HMM, as a dynamic time-series statistical model, has the reliable computing performance and rigorous data structure, and has been widely used for speech recognition [1,2]. While SVM with a solid mathematical theoretical basis is a kind of high-performance method to express the dependence of complex functions in the high-dimensional space.…”

Section: Introductionmentioning

confidence: 99%

State Monitoring for Centrifugal Pump of PWR Based on HMM and SVM

Zhang

Xia

2010

AMR

View full text Add to dashboard Cite

The centrifugal pump of pressurized water reactor (PWR) in nuclear power plant is characterized by its complicated system, small accumulated data and fault samples. HMM has a strong ability to deal with time series modeling for dynamic process, while SVM has excellent generalization ability to solve the learning problems with small samples. This paper develops a state monitoring system based on the hybrid HMM/SVM model. The wavelet analysis techniques are used to extract features and the Hidden Markov Model (HMM) and Support Vector Machine (SVM) are used as the basic modeling and identification tools. The identification results of centrifugal pump show that the hybrid HMM/SVM system is effective and available for the state monitoring of the centrifugal pump of PWR in nuclear power plan.

show abstract

An integrated study of speaker normalisation and HMM adaptation for noise robust speaker-independent speech recognition

Cited by 3 publications

References 14 publications

Acoustic Model Adaptation for Indonesian Language Utterance Training System

Acoustic Model Adaptation for Indonesian Language Utterance Training System

Study of On-Line Condition Monitoring System for Roller Based on HMM

State Monitoring for Centrifugal Pump of PWR Based on HMM and SVM

Contact Info

Product

Resources

About