Understanding how Deep Belief Networks perform acoustic modelling

By Abdel-rahman Mohamed, Geoffrey E. Hinton, Gerald Penn. 2012

322 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Probabilistic & Bayesian Methods, Speech & Audio, 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 Probabilistic & Bayesian Methods

In Speech & Audio

In Signal Processing

Citations in the corpus

Effectiveness of self-supervised pre-training for speech recognition (2019, neutral): “Similar to the observations in (Mohamed et al., 2012), the FBANK features are more friendly to unsupervised local acoustic representation learning methods like continuous BERT, leading to consistent gains compared to MFCC features for both 10h and 1h sets.”
Deep Learning: Methods and Applications (2014, methodology): “For example, [259] applied a dimensionality reduction method to visualize the relationship among the feature vectors learned by the DNN.”
Analyzing Hidden Representations in End-to-End Automatic Speech Recognition Systems (2017, background): “For instance, previous work evaluated neural network acoustic models on phoneme recognition using different acoustic features [11] or investigated how such models learn invariant representations [12] and encode linguistic features [13, 14].”
A tutorial survey of architectures, algorithms, and applications for deep learning (2014, background): “An insightful analysis on what speech information DBNs can capture is provided in [26].”
Encoding of phonology in a recurrent neural model of grounded speech (2017, methodology): “Mohamed et al. (2012) use a Deep Belief Network (DBN) for acoustic modeling and phone recognition on human speech.”
Recent progresses in deep learning based acoustic models (2017, methodology): “Typically, log Mel-ﬁlter-bank features are often used as the input to the neural-network-based acoustic model [33], [34].”
Visualizing the Hidden Activity of Artificial Neural Networks (2017, methodology): “5, have been used as high-level evidence of model efficacy [7, 29, 41, 36, 16].”
SpeechMix - Augmenting Deep Sound Recognition Using Hidden Space Interpolations (2020, background): “We could think of lower level activations of a DNN as speaker-adapted features [29], while the upper layer activations could be thought of as performing class-based discrimination.”
A Review of Deep Learning with Special Emphasis on Architectures, Applications and Recent Trends (2019, methodology): “…transformation between two input images, Convolutional RBM [60], [61] to understand the time structure of the input time series, mean-covariance RBM [62], [63], [64] to represent the covariance structure of the data, and many more like Recurrent TRBM [65], factored conditional RBM (fcRBM) [66].”
Extracting deep bottleneck features using stacked auto-encoders (2013, background): “In our experiments, we extracted deep bottleneck features from both MFCCs and log mel scale filterbank coefficients (lMEL).”
Elastic spectral distortion for low resource speech recognition with deep neural networks (2013, methodology): “In our experiments, logMFB features always produced better results than MFCC features, the same as mentioned in the paper [20].”
Multitask Learning with Low-Level Auxiliary Tasks for Encoder-Decoder Based Speech Recognition (2017, background): “This effect has been found in systems for speech processing [14, 15] as well as computer vision [16, 17].”
An Overview of Noise-Robust Automatic Speech Recognition (2014, methodology): “In [78], [291], it is shown that log-ﬁ lter-bank features are significantly better than the widely-used MFCCs when used as the input to the DNN.”
Multichannel Signal Processing With Deep Neural Networks for Automatic Speech Recognition (2017, methodology): “DNNs are attractive because they have been shown to be able to perform feature extraction jointly with classification [9].”
Discriminatively trained recurrent neural networks for single-channel speech separation (2014, result): “Furthermore, it is notable that the choice of feature representation has such a strong effect on the results, but this is in accordance with earlier studies showing that DNN acoustic models cannot compensate even for simple rotations of the input features [26].”
Deep Neural Networks for Acoustic Modeling in Speech Recognition (2012, methodology): “Subsequent work combined the two approaches by using state-of-the-art, discriminative ly trained (DT) speaker-dependent features as input to the DBN-DNN [24], but this produced little further improvement , probably because the hidden layers of the DBN-DNN were already doing quite a good job of progr”
On internal language representations in deep learning: an analysis of machine translation and speech recognition (2018, background): “For instance, previous work evaluated neural network acoustic models on phone recognition using different acoustic features [243] or investigated how such models learn invariant representations [352] and encode linguistic features in different layers [251, 252].”