Adaptation of polynomial trajectory segment models for large vocabulary speech recognition

Kannan, A.; Ostendorf, Mari

doi:10.1109/icassp.1997.596212

Cited by 10 publications

(20 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The clustering algorithm for PSM's has applications beyond providing robust estimates for context-models, in that it can be used for defining equivalence classes for adaptation parameter tying [16] as previously proposed for HMM's [17], [18]. In addition, the clustering algorithm described here can be used with segment models having hidden polynomial trajectories [19] by first clustering assuming a deterministic trajectory and then estimating the Gaussian prior for the mean trajectory parameters based on data assigned to each cluster.…”

Section: Discussionmentioning

confidence: 99%

A comparison of constrained trajectory segment models for large vocabulary speech recognition

Kannan

Ostendorf²

1998

IEEE Trans. Speech Audio Process.

Self Cite

View full text Add to dashboard Cite

This paper compares parametric and nonparametric constrained-mean trajectory segment models for large vocabulary speech recognition, extending distribution clustering techniques to handle polynomial mean trajectory models for robust parameter estimation. The parametric model has fewer free parameters and gives similar recognition performance to the nonparametric model, but has higher recognition costs.Index Terms-Distribution clustering, polynomial mean trajectory, segment model.

show abstract

Section: Discussionmentioning

confidence: 99%

A comparison of constrained trajectory segment models for large vocabulary speech recognition

Kannan

Ostendorf²

1998

IEEE Trans. Speech Audio Process.

Self Cite

View full text Add to dashboard Cite

show abstract

“…The formulation of the trajectory-based HMM or nonstationary-state HMM has been successfully used in automatic speech recognition applications for the past few years [3], [5]. The trended HMM is of a data-generative type and can be described as…”

Section: Parameter Estimation Of Linear Transformation Matricesmentioning

confidence: 99%

“…Based on the model , the optimum state sequence for an input token with frames is obtained by means of the Viterbi algorithm [3]. Then, the log-likelihood is given by (5) In this section, the discriminative training process is briefly summarized for achieving optimal accuracy in estimating the state-dependent transformation matrix coefficients. Let , , denote the HMM for the th class, where is the total number of classes.…”

Section: Parameter Estimation Of Linear Transformation Matricesmentioning

confidence: 99%

Linear trajectory models incorporating preprocessing parameters for speech recognition

Chengalvarayan¹

1998

IEEE Signal Process. Lett.

View full text Add to dashboard Cite

In this letter, we investigate the interactions of frontend feature extraction and back-end classification techniques in nonstationary state hidden Markov model (NSHMM) based speech recognition. The proposed model aims at finding an optimal linear transformation on the mel-warped discrete Fourier transform (DFT) features according to the minimum classification error (MCE) criterion. This linear transformation, along with the NSHMM parameters, are automatically trained using the gradient descent method. An error rate reduction of 8% is obtained on a standard 39-class TIMIT phone classification task in comparison with the MCE-trained NSHMM using conventional preprocessing techniques.Index Terms-Discrete cosine transform, discriminative feature extraction, gradient descent algorithm, hidden Markov model, mel-warped discrete Fourier transform.

show abstract

“…In [10], the PSM-based MAP adaptation was proposed while in [11], we proposed the PSM-based MLLR adaptation. For very small amount of adaptation data, or for very fast adaptation, EV or RSW adaptations are needed.…”

Section: Psm-based Rsw Adaptationmentioning

confidence: 99%

“…In [10], PSM-based MAP adaptation was proposed. In this paper, we focus on the task of rapid adaptation and derive the PSM-based RSW algorithm.…”

Section: Introductionmentioning

confidence: 99%

Reference Speaker Weighting Adaptation for Sub-Phonetic Polynomial Segment Models

Yeung

Siu

2006 IEEE International Conference on Acoustics Speed and Signal Processing Proceedings

View full text Add to dashboard Cite

Speaker adaptation has been widely used in speech recognition. With small amount of adaptation data, Reference Speaker Weighting (RSW) adaptation was previously proposed for fast HMM adaptation, and has been shown to outperform the more commonly used maximum likelihood linear regression (MLLR) adaptation. Extending our previous work [1, 2] of applying the Polynomial Segment Models (PSMs) in large vocabulary continuous speech recognition (LVCSR) on the WSJ Nov 92 evaluation, we derive the PSM-based RSW fast adaptation technique in this paper. Different from the HMMs, in which the model means are constants within a state, the PSM means are curves represented by polynomials. Experimental results showed that the PSM-based RSW gave approximately the same relative improvement over the unadapted model as in the HMM case. Comparing the PSM-based RSW and MLLR, the PSM-based RSW is more powerful when the amount of adaptation data available is limited. However, it could quickly saturate with increase in adaptation data.

show abstract

Adaptation of polynomial trajectory segment models for large vocabulary speech recognition

Cited by 10 publications

References 9 publications

A comparison of constrained trajectory segment models for large vocabulary speech recognition

A comparison of constrained trajectory segment models for large vocabulary speech recognition

Linear trajectory models incorporating preprocessing parameters for speech recognition

Reference Speaker Weighting Adaptation for Sub-Phonetic Polynomial Segment Models

Contact Info

Product

Resources

About