Connectionist probability estimators in HMM speech recognition

By S. Renals, Nelson Morgan, Hervé Bourlard, Michael Cohen, Horacio Franco. 1994

Subfields: Probabilistic & Bayesian Methods, Speech & Audio

How researchers see this paper

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

In Probabilistic & Bayesian Methods

In Speech & Audio

Citations in the corpus

Deep Speech 2 : End-to-End Speech Recognition in English and Mandarin (2015, methodology): “Recurrent neural networks and networks with convolution were also used in speech recognition around the same time [51, 67].”
Deep Speech: Scaling up end-to-end speech recognition (2014, background): “Neural network acoustic models and other connectionist approaches were ﬁrst introduced to speech pipelines in the early 1990s [1, 34, 11].”
Effective Training of a Neural Network Character Classifier for Word Recognition (1996, background): “…work exists to indicate the general applicability of ANN technology as classifiers providing good estimates of a posteriori probabilities of each class given the input (Gish 1990, Richard and Lippmann 1991, Renals and Morgan 1992, Lippmann 1994, Morgan and Bourlard 1995, and others cited herein).”
From HMM's to segment models: a unified view of stochastic modeling for speech recognition (1996, background): “[85, 86, 87, 88]), and under certain assumptions can be shown to be mathematically consistent [89].”
Large vocabulary continuous speech recognition with context-dependent DBN-HMMS (2011, unknown): “Index Terms— Speech recognition, deep belief network, context-dependent phone, LVCSR, DBN-HMM”
Multilingual training of deep neural networks (2013, methodology): “While the basic idea was used in speech recognition in the early 1990s [15], earlier uses of hybrid systems were mainly limited to estimating scaled likelihoods for monophone states using feedforward network with two layers [21] and recurrent networks [22], owing to computational constraints.”
Large Vocabulary Speech Recognition Using Deep Tensor Neural Networks (2012, methodology): “…deep neural network hidden Markov models (CD-DNN-HMMs) for large vocabulary speech recognition and achieved highly promising recognition results including over one third fewer word errors than the discriminatively trained, conventional HMM-based systems on the 300hr Switchboard benchmark task.”
Hybrid acoustic models for distant and multichannel large vocabulary speech recognition (2013, methodology): “In a deep neural network-hidden Markov model hybrid system [19, 20, 16], the DNN is trained to classify the input acoustics into classes corresponding to the HMM states.”
Pipelined BackPropagation for Context-Dependent Deep Neural Networks (2012, methodology): “Like the ANN-HMMs of the 90’s [4], CD-DNN-HMMs replace GMMs by an artificial neural network, but they differ in significantly increased network depth (7 or more hidden layers) and in that theydirectlymodel tiedcontext-dependent s ates (senones) [5, 2] instead of factorizing the networks [6, 7].”
Roles of Pre-Training and Fine-Tuning in Context-Dependent DBN-HMMs for Real-World Speech Recognition (2010, methodology): “The key difference (excluding the use of DBN pre-training) between the CD-DBN-HMM architecture and the earlier CD-ANN-HMM architecture [3] is that we use senones as the DBN output units directly.”
Multi-level adaptive networks in tandem and hybrid ASR systems (2013, methodology): “In an ASR system, neural networks may be used to directly compute HMM observation probabilities (the hybrid approach [1, 2]) or for feature extraction (the tandem approach [3]).”
An application of recurrent nets to phone probability estimation (1994, methodology): “For a full discussion of this result to speech recognition see [18, 19].”
On Speech Recognition (2012, methodology): “In particular the author look at some of the technical developments underpinning these recent developments and look ahead to current study which promises to enable the next wave of innovations in accuracy and scale for speech processing.”
Audio-visual speech recognition using deep learning (2014, methodology): “The idea of utilizing a neural network to replace a GMM and construct a hybrid model that combines a multilayer perceptron and HMMs was originally proposed decades ago [8, 41].”