Home HomeABSTRACT
Inconsistency between training and testing criteria is a drawback of the hybrid artifcial neural network and hidden Markov model (ANN/HMM) approach to speech recognition. This paper presents an effective method to address this problem by modifying the feedforward neural network training paradigm. Word errors are explicitly incorporated in the training procedure to achieve improved word recognition accuracy. Experiments on a continuous digit database show a reduction in word error rate of more than 17% using the proposed method.
A0108.pdf
TOPABSTRACT
In this paper, hybrid HMM/ANN systems are used to model context dependent phones. In order to reduce the number of parameters as well as to better catch the dynamics of the phonetic segments, we combine (context dependent) diphone models with context independent phone models. Transitions from phone to phone are modeled as generalized context dependent distributions while phonetic units are context independent models trained on the less coarticulated middle part of each phone. Words are thus modeled as a sequence of probability distributions alternatively representing the middle part of the phonemes and the transitions from phone to phone. A single neural network is used to estimate both context independent phone probabilities and generalized context dependent diphone (phone to phone transition) probabilities. Resulting systems are compared to classical context independent phone-based HMM/ANN systems with the same number of parameters. The Phonebook isolated word database has been used for training the systems. Testing is done on small (75 words), medium (600 words) and large (8000 words) lexicons. Test words were not present in the training vocabulary.
A0213.pdf
TOPABSTRACT
The results of our research presented in this paper are 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- back- ward hybrid systems are presented and discussed.
A0233.pdf
TOPABSTRACT
In this paper we introduce four acoustic confidence measures which are derived from the output of a hybrid HMM/ANN large vocabulary continuous speech recognition system. These confidence measures, based on local posterior probability estimates computed by an ANN, are evaluated at both phone and word levels, using the North American Business News corpus.
A0237.pdf
TOPABSTRACT
In this paper we investigate a number of ensemble methods for improving the performance of connectionist acoustic models for large vocabulary continuous speech recognition. We discuss boosting, a data selection technique which results in an ensemble of models, and mixtures-of- experts. These techniques have been applied to multi- layer perceptron acoustic models used to build a hy- brid connectionist-HMM speech recognition system. We present results on a number of ARPA benchmark tasks, and show that the ensemble methods lead to considerable improvements in recognition accuracy.
A0400.pdf
TOPABSTRACT
This paper presents results of our efforts on combining standard mixture of Gaussians acoustic modeling [10] with a context-dependent hybrid connectionist HME/HMM architecture [3, 4] for the Switchboard corpus. Using a score normalization scheme which is independent of the stream's modeling paradigm and adaptive methods for combining multiple probability distributions, we achieve a relative decrease in word error rate of 3.5% and 9.3%, compared to each of the single stream systems. As opposed to multiple acoustic streams based on mixture of Gaussians, the integration of hybrid NN/HMM based modeling appears to be advantageous since the differences in modeling techniques and training algorithms allow to capture different aspects of the speech signal. Small dependence among emission probability estimates is considered essential for potential gains in interpolated systems.
A0931.pdf
TOP