Large scale discriminative training of hidden Markov models for speech recognition

By Philip C. Woodland, Daniel Povey. Computer Speech and Language, 2002

369 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Signal Processing

View Paper: https://www.semanticscholar.org/paper/2a6ae3d667a5c2601c1852a0753c8b1c749fec1e

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Signal Processing

Citations in the corpus

Learning from past mistakes: improving automatic speech recognition output via noisy-clean phrase context modeling (2018, methodology): “Further, our work is different from discriminative training of acoustic [33] models and discriminative language models (DLM) [34], which are trained directly to optimize the word error rate using the reference transcripts.”
Error Approximation and Minimum Phone Error Acoustic Model Estimation (2010, methodology): “competing state sequences have larger posterior probabilities and it has be en observed [2] that use of acoustic scaling leads to improved generalisation of discriminatively-estimated acoustic m odels.”
Hypothesis spaces for minimum Bayes risk training in large vocabulary speech recognition (2006, methodology): “Practical solutions to this problem are to approximate this space either using an N-best list of the most likely hypotheses [2] or to use lattices as a more compact representation [1].”
Semi-supervised maximum mutual information training of deep neural network acoustic models (2015, methodology): “A number of discriminative criteria such as MMI [1], MCE[2], MPE [3], sMBR [4, 5] and bMMI [6] have been developed and used in HMM-based speech recognition [7, 8, 9, 10].”
Unsupervised Adaptation With Discriminative Mapping Transforms (2009, methodology): “Once the DLTs are estimated, the form of model adaptation remains the same (4) where is the DLT of speaker .”
Minimum Phone Error and I-smoothing for improved discriminative training (2002, methodology): “Further details of this training corpus, denoted h5train00, are given in [11].”
Mutual Information and Diverse Decoding Improve Neural Machine Translation (2016, background): “Neural machine translation models map source x = { x 1 , x 2 , ...x N x } to a continuous vector representation, from which target output y = { y 1 , y 2 , ..., y N y } is to be generated.”
Unsupervised Testing Strategies for ASR (2011, methodology): “To improve the generalization of discriminative training, we use a unigram to describe the space of potential errors [7], but a trigram or higher to give us transcription truth with unsupervised training.”
Parameter Estimation of Statistical Models Using Convex Optimization (2010, methodology): “The essential idea of DT is to apply a discriminative criterion to the training procedure of HMMs.”
Machine Learning Paradigms for Speech Recognition: An Overview (2013, methodology): “Due to its discriminative nature, MMIE has demonstrated significant performance improvement over using the joint likelihood loss in training Gaussian-mixture HMM systems [99]–[101].”
Neural conditional random fields (2010, methodology): “In the following, we only considered NeuroCRFs trained with the CML criterion.”
Optimization Algorithms and Applications for Speech and Language Processing (2013, methodology): “The introduction of discriminative training for HMMs [4], [76]–[78] raised the question of whether there exist lower bounds for the discriminative formulation as well.”
End-to-end Speech Recognition Using Lattice-free MMI (2018, methodology): “The denominator graph has traditionally been estimated using n-best lists and later using lattices [21][22].”