Probabilistic and Bottle-Neck Features for LVCSR of Meetings

By Frantisek Grézl, Martin Karafiát, S. Kontar, Jan Honza Cernocký. IEEE International Conference on Acoustics, Speech, and Signal Processing, 2007

390 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Signal Processing

View Paper: https://www.semanticscholar.org/paper/5fc7204b78c4d85e9d59496fcc067d62f6f4a7e1

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

End-to-End attention based text-dependent speaker verification (2016, background): “layer of the DNN before the sigmoid nonlinearity [26].”
Investigation on cross- and multilingual MLP features under matched and mismatched acoustical conditions (2013, methodology): “Beside the hybrid acoustic modeling [2], the MLP based posterior estimations can be integrated into the Gaussian Mixture based Hidden Markov Model (GMM-HMM) framework through either probabilistic or bottleneck TANDEM approach [3, 4].”
Hybrid acoustic models for distant and multichannel large vocabulary speech recognition (2013, methodology): “Most previous research using the AMI corpus [3, 22] have done so in the context of the NIST RT evaluations, where the AMI data was used together with other meeting corpora.”
Investigation of unsupervised adaptation of DNN acoustic models with filter bank input (2014, methodology): “To perform nonlinear feature compression, the first DNN has a bottleneck (BN) layer [7], which consists of a limited number of units.”
Deep neural network features and semi-supervised training for low resource speech recognition (2013, background): “Both posterior and bottleneck features [23, 24] can be derived from the DNN.”
Attention-based Wav2Text with feature transfer learning (2017, neutral): “One popular scheme for utilizing DNNs for transfer learning within ASR frameworks is a tandem approach [29].”
Hierarchical Bayesian combination of plug-in maximum a posteriori decoders in deep neural networks-based speech recognition and speaker adaptation (2017, methodology): “The SMAP (Shinoda and Lee, 2001) technique is used to perform speaker adaptation within BN-CD-GMM-HMMs.”
Learning hidden unit contributions for unsupervised speaker adaptation of neural network acoustic models (2014, methodology): “These results use neural networks as both hybrid systems [2,3] where the neural network provides a scaled likelihood estimate which replaces the GMM, and as tandem/bottleneck systems [4, 5] in which the neural network is used as a discriminative feature extractor for a GMM-based system.”
The language-independent bottleneck features (2012, background): “In the experiments with GlobalPhone database, we show that Multilingual BN-features consistently outperform Monolingual BN-features.”
Extracting deep neural network bottleneck features using low-rank matrix factorization (2014, methodology): “This is done by training a network to predict phonetic targets, and then either using the estimated target probabilities [1] or the activations of a narrow hidden “bottleneck” (BN) layer [2] as features for a standard GMM-HMM configuration.”
Gate Activation Signal Analysis for Gated Recurrent Neural Networks and its Correlation with Phoneme Boundaries (2017, background): “A good example is the bottleneck features [8][9].”
Multilingual MLP features for low-resource LVCSR systems (2012, background): “Bottleneck features - Unlike Tandem features, bottleneck features are derived as linear outputs of the neurons from the bottleneck layer [6].”
Speech Recognition Using Long-Span Temporal Patterns in a Deep Network Model (2013, background): “Bottle-Neck features, first introduced in [8], are an extension of these TANDEM features, and allow better recognition results.”
Using neural network front-ends on far field multiple microphones based speech recognition (2014, background): “Index Terms— speech recognition, multiple microphone, beamforming, deep neural networks”
Sparse Multilayer Perceptron for Phoneme Recognition (2012, methodology): “Alternatively, discriminative features derived from the hidden layer of an MLP [23], [24] are used as features in large vocabulary continuous speech recognition [25], [26].”