REMAP: recursive estimation and maximization of a posteriori probabilities in connectionist speech recognition

Bourlard, Hervé; Konig, Yochai; Morgan, Nelson

doi:10.21437/eurospeech.1995-402

Cited by 10 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this paper: (1) we demonstrate that the original HMM/ANN system 3, 4] trained using local criteria indeed optimizes the global posterior probability, given certain well-de ned assumptions (2) we use the REMAP algorithm to derive a f o r w ard-backward training algorithm for the original HMM/ANN system (3) we demonstrate the performance of these algorithms on the task-independent Phonebook database.…”

Section: Introductionmentioning

confidence: 90%

Estimation of global posteriors and forward-backward training of hybrid HMM/ANN systems

Hennebert¹,

Ris²,

Bourlard³

et al. 1997

5th European Conference on Speech Communication and Technology (Eurospeech 1997)

View full text Add to dashboard Cite

The results of our research presented in this paper is two-fold. First, an estimation of global posteriors is formalized in the framework of hybrid HMM/ANN systems. It is shown that hybrid HMM/ANN systems, in which the ANN part estimates local posteriors, can be used to modelize global model posteriors. This formalization provides us with a clear theory in which both REMAP and \classical" Viterbi trained hybrid systems are uni ed. Second, a new forward-backward training of hybrid HMM/ANN systems is derived from the previous formulation. Comparisons of performance between Viterbi and forward-backward hybrid systems are presented and discussed.

show abstract

Section: Introductionmentioning

confidence: 90%

Estimation of global posteriors and forward-backward training of hybrid HMM/ANN systems

Hennebert¹,

Ris²,

Bourlard³

et al. 1997

5th European Conference on Speech Communication and Technology (Eurospeech 1997)

View full text Add to dashboard Cite

show abstract

“…A Markov model output denes an a posteriori probability P (wjx), for every observation sequence x and sentence transcription w. All model parameters can thus be optimised with the global discriminative criterion E = X n log P (w n jx n ); (1) where n is the trainset index. This criterion is known as Conditional Maximum Likelihood (CML) as proposed by Brown in [13], and has also been referred to as the global MAP criterion [14,12]. When the language model P (w) i s k ept constant, CML/MAP is also equivalent to Maximum Mutual Information (MMI) training.…”

Section: Training and Decodingmentioning

confidence: 99%

“…ANN-based hybrids were originally trained to do single frame discrimination by the embedded Viterbi algorithm [9,10]. Global discriminative training of ANN-based systems has however also been proposed [11,12].…”

Section: Introductionmentioning

confidence: 99%

A comparison of hybrid HMM architectures using global discriminative training

Johansen¹

1996

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

View full text Add to dashboard Cite

This paper presents a comparison of dierent model architectures for TIMIT phoneme recognition. The baseline is a conventional diagonal covariance Gaussian mixture HMM. This system is compared to two dierent h ybrid MLP/HMMs, both adhering to the same restrictions regarding input context and output states as the Gaussian mixtures. All free parameters in the three systems are jointly optimised using the same global discriminative criterion. A Forward decoder, with total likelihood scoring, is used for recognition. While the global discriminative training method is found to improve the baseline HMM signicantly, the dierences between Gaussian and MLP-based architectures are small. The Gaussian mixture system however performs slightly better at the lowest complexity levels.

show abstract

Bidirectional LSTM Networks for Improved Phoneme Classification and Recognition

Graves

Fernández

Schmidhuber

2005

Lecture Notes in Computer Science

595

253

View full text Add to dashboard Cite

In this paper, we carry out two experiments on the TIMIT speech corpus with bidirectional and unidirectional Long Short Term Memory (LSTM) networks. In the first experiment (framewise phoneme classification) we find that bidirectional LSTM outperforms both unidirectional LSTM and conventional Recurrent Neural Networks (RNNs). In the second (phoneme recognition) we find that a hybrid BLSTM-HMM system improves on an equivalent traditional HMM system, as well as unidirectional LSTM-HMM. 2 LSTM LSTM [9,6] is an RNN architecture designed to deal with long time-dependencies. It was motivated by an analysis of error flow in existing RNNs [8], which found that long

show abstract

REMAP: recursive estimation and maximization of a posteriori probabilities in connectionist speech recognition

Cited by 10 publications

References 0 publications

Estimation of global posteriors and forward-backward training of hybrid HMM/ANN systems

Estimation of global posteriors and forward-backward training of hybrid HMM/ANN systems

A comparison of hybrid HMM architectures using global discriminative training

Bidirectional LSTM Networks for Improved Phoneme Classification and Recognition

Contact Info

Product

Resources

About