Minimum Phone Error and I-smoothing for improved discriminative training

By Daniel Povey, Philip C. Woodland. 2002

811 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Signal Processing

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 Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Deep Learning: Methods and Applications (2014, methodology): “Here, it is useful to point out a connection between the above pretraining/fine-tuning strategy associated with hybrid deep networks and the highly popular minimum phone error (MPE) training technique for the HMM (see [147, 290] for an overview).”
Parallel training of Deep Neural Networks with Natural Gradient and Parameter Averaging (2014, methodology): “We generally use state-level Minimum Bayes Risk (sMBR) [27, 28] although we have also implemented Minimum Phone Error (MPE) [26] and Boosted MMI [25].”
Feature Learning in Deep Neural Networks - A Study on Speech Recognition Tasks (2013, methodology): “The model parameters can be discriminatively trained using an objective function such as maximum mutual information (MMI) [1] or minimum phone error rate (MPE) [2].”
A tutorial survey of architectures, algorithms, and applications for deep learning (2014, methodology): “Here, it is useful to point out a connection between the above hybrid discriminative training and a highly popular MPE (minimum phone error) training technique for the HMM (Povey and Woodland, 2002).”
Spoken Content Retrieval—Beyond Cascading Speech Recognition with Text Retrieval (2015, methodology): “1) Weighted Discriminative Training: Discriminative training techniques such as minimum classification error (MCE) [113] and minimum phone error (MPE) [30], [114] training have been widely used to obtain better HMM acoustic”
Unsupervised Adaptation With Discriminative Mapping Transforms (2009, methodology): “The occupan cy γ m(ts) and γ d m(ts) are calculated using the lattice forwardbackward algorithm [21] andγ m (ts) is the normal ML posterior occupancy calculated given the correct transcripti on; α is a smoothing constant to balance the ML occupancy and the other occupancy to improve the generalisati”
Cross-Domain and Cross-Language Portability of Acoustic Features Estimated by Multilayer Perceptrons (2006, methodology): “All acoustic model s are trained discriminatively using the minimum phone error (MPE) criterion [16].”
Multi-level adaptive networks in tandem and hybrid ASR systems (2013, methodology): “However, it should be noted that the tandem system beneﬁts from more powerful adaptation transforms and the use of MPE training.”
Unsupervised training and directed manual transcription for LVCSR (2010, methodology): “These approximate transcriptions can be used with lightly-supervised training (Lamel et al., 2002; Chan and Woodland, 2004).”
Which words are hard to recognize? Prosodic, lexical, and disfluency factors that increase speech recognition error rates (2010, methodology): “Both systems have gender-dependent acoustic models trained discriminatively using variants of minimum phone error (MPE) training with maximum mutual information (MMI) priors (Povey and Woodland, 2002).”
Lattice-based optimization of sequence classification criteria for neural-network acoustic modeling (2009, methodology): “Other studies [6, 7] demonstrate the effectiveness of global normalization and conditional maximum likelihood (CML) training (equivalent to MMI for a fixed language model) on TIMIT phone recognition and broad class recognition.”
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].”
Analysis of MLP-Based Hierarchical Phoneme Posterior Probability Estimator (2011, methodology): “The linear transformation matrix is trained u sing minimum phone error criterion [37].”
Text-to-Speech Synthesis (2009, methodology): “More recently, discriminative training has become the norm in ASR [495], [360] where HMMs as described are used, but where their parameters are trained to maximise discriminat io , not data likelihood.”
Boosted MMI for model and feature-space discriminative training (2008, background): “This raw accuracy must be approximated for efficiency [6].”
Speech recognition and keyword spotting for low-resource languages: Babel project research at CUED (2014, methodology): “Using these unsupervised transcriptions for discriminative training (either MPE or fMPE) degraded performance 3.”
Scalable Minimum Bayes Risk Training of Deep Neural Network Acoustic Models Using Distributed Hessian-free Optimization (2012, background): “In contrast, sequence-discriminative criteria such as maximum mutual information (MMI) [4] and minimum Bayes risk [5, 6] have become the norm for Gaussian mixture model (GMM) acoustic models.”
Hidden conditional random fields for phone classification (2005, background): “We note that this approach is generalizable to other smooth discriminative criteria such as the conditional expectation of the raw phone or word error rate [10], or the smoothed empirical error of the training data [11].”