The acoustic-modeling problem in automatic speech recognition

By Peter F. Brown. 1987

Subfields: Natural Language Processing, 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 Natural Language Processing

In Speech & Audio

Citations in the corpus

A Diversity-Promoting Objective Function for Neural Conversation Models (2015, methodology): “speech recognition (Bahl et al., 1986; Brown, 1987), as an optimization objective that measures the mutual dependence between inputs and outputs.”
A Survey on Dialogue Systems: Recent Advances and New Frontiers (2017, methodology): “They used a Maximum Mutual Information (MMI), which was ﬁrst introduced in speech recognition [6; 9], as an optimization objective.”
Audio Chord Recognition with Recurrent Neural Networks (2013, background): “As argued previously [8], the relatively poor performance of the HMM may be due to the context information added to the input x in equations (3-4).”
A time-delay neural network architecture for isolated word recognition (1990, methodology): “Versions of back-propagation were independently derived in Parker (1985) and Werbos (1974). sessed both of the crucial featur~:,~ of Brown~ irisproved hidden Markov model: the input to the system consisted of vectors of real-valued acoustic parameters, and the training algorithm explicitly caused di”
Acoustic Modeling Using Deep Belief Networks (2012, background): “These approaches coincided ni cely with a trend initiated by [9] in which generative modeling is replaced by discriminative training.”
Reformulating the HMM as a trajectory model by imposing explicit relationships between static and dynamic feature vector sequences (2004, unknown): “, 2003), frame-correlated HMM (Wellekens, 1987; Brown, 1987; Takahashi, 1993; Paliwal, 1993; Kim and Un, 1997; Qing et al., 1999), partly hidden Markov model (Kobayashi et al.”
Markovian Models for Sequential Data (2004, background): “The most common are the maximum a posteriori criterion, to maximize P (w 1 jy T 1 ), and the maximum mutual information criterion [77, 78, 79], to maximize log P (y 1 jw L 1 ) P (y 1 ) .”
Using Generative Models for Handwritten Digit Recognition (1996, background): “…than correct modeling of the image data, and it generally leads to better discriminative performance [42], although the advantage of discriminative learning vanishes if the generative model is correct and the (cid:12)tting process produces the true probability of the data given the model [43].”
Dimensionality Reduction and Prior Knowledge in E-Set Recognition (1989, methodology): “In fact, the TDNN is probably the best automatic recognition system built for this task to date; it even performs slightly better than the continuous acoustic parameter, maximum mutual information hidden Markov model proposed in (Brown, 1987).”
Phoneme recognition using time-delay neural networks (1989, background): “Excellent performance was achieved at all levels from the phonemic level 1351-[38] to word recognition [39], [34] and to continuous speech recognition 1401.”