Exemplar-Based Sparse Representation Features: From TIMIT to LVCSR

By Tara N. Sainath, Bhuvana Ramabhadran, Michael A. Picheny, David Nahamoo, Dimitri Kanevsky. 2011

75 citations recorded by the source corpus.

How researchers see this paper

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

Citations in the corpus

Deep Learning: Methods and Applications (2014, methodology): “Further, the DNN results were found to be slightly superior to the then-best-performing single system based on the generative hidden trajectory model (HTM) in the literature [105, 108] evaluated on the same, commonly used TIMIT task by many speech researchers [107, 108, 274, 313].”
TuPAQ: An Efficient Planner for Large-scale Predictive Analytic Queries (2015, methodology): “The second dataset is a pre-featurized version of the TIMIT Acoustic-Phonetic continuous speech corpus [24], featurized according to the procedure described in [44]—yielding roughly 2 , 300 , 000 examples each having 440 features.”
Learning filter banks within a deep neural network framework (2013, background): “For example, in large vocabulary continuous speech recognition systems, huge gains in performance are observed by using speaker-adapted and discriminatively trained-features [1], learned via objective functions such as feature-space maximum likelihood linear regression (fMLLR), and feature-space boo”
Machine Learning Paradigms for Speech Recognition: An Overview (2013, methodology): “Such sparse representations are used as additional features to the existing high-quality features and error rate reduction is reported in both phone recognition and large vocabulary continuous speech recognition tasks with detailed experimental conditions provided in [263].”
Automating model search for large scale machine learning (2015, methodology): “The second dataset is a pre-featurized version of the TIMIT Acoustic-Phonetic continuous speech corpus [20], featurized according to the procedure described in [42]—yielding roughly 2,300,000 examples each having 440 features.”
Deep Neural Networks for Acoustic Modeling in Speech Recognition (2012, methodology): “Using contextdependent HMM states, it is possible to outperform state-of-theart BMMI trained GMM-HMM systems with a two-hidden-layer neural network without using any pretraining [43], though using more hidden layers and pretraining works even better.”
Convolutional, Long Short-Term Memory, fully connected Deep Neural Networks (2015, methodology): “This recipe order has been shown to be appropriate for LVCSR tasks [7].”
Deep Belief Networks using discriminative features for phone recognition (2011, methodology): “On TIMIT, we found that these features 2The GMM system is the best number on TIMIT with speaker adapted , discriminative features [16].”
Deep Scattering Spectrum with deep neural networks (2014, methodology): “We explore the behavior of scattering features using 2,400 CD-state output targets [17].”
An Overview of Noise-Robust Automatic Speech Recognition (2014, background): “Recently, there is increasing interest to use exemplar-based methods for general ASR [161], [162] and noise-robust ASR [163]–[165].”