Deep Scattering Spectrum with deep neural networks

Peddinti, Vijayaditya; Sainath, Tara N.; Maymon, Shay; Ramabhadran, Bhuvana; Nahamoo, D.; Goel, Vaibhava

doi:10.1109/icassp.2014.6853588

Cited by 31 publications

(22 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The main challenge in incorporating second order information into speech recognition systems is in that first order coefficients typically require some further band-pass filtering to capture local invariants in such spectro-temporal decompositions of speech waveforms. Empirically, the most effective neural architecture for hybrid acoustic models with scattering features has been proposed in [19]. The architecture is a junction network that takes as input first and second order scattering coefficients via separate pipelines which are then merged into a multi-layer perceptron with several hidden layers.…”

Section: Network Architecturementioning

confidence: 99%

“…The pooling is applied with compression rates 3, 2 and 1, respectively. We note here that our configuration of channels and filter sizes is different from prior work [10,19]. The pipeline for second order coefficients is a multi-layer perceptron (MLP) with 512 activation units, followed by batch normalization [22], RELU non-linearity, and a dropout block.…”

Section: Network Architecturementioning

confidence: 99%

“…In the first experiment, the goal is to establish that second order scattering coefficients contain invariants relevant for robust- are processed using a convolutional neural network, while the combination of first and second order features employs the junction architecture from [19].…”

Section: Clean Condition Trainingmentioning

confidence: 99%

See 2 more Smart Citations

Deep Scattering Power Spectrum Features for Robust Speech Recognition

et al. 2020

View full text Add to dashboard Cite

Deep scattering spectrum consists of a cascade of wavelet transforms and modulus non-linearity. It generates features of different orders, with the first order coefficients approximately equal to the Mel-frequency cepstrum, and higher order coefficients recovering information lost at lower levels. We investigate the effect of including the information recovered by higher order coefficients on the robustness of speech recognition. To that end, we also propose a modification to the original scattering transform tailored for noisy speech. In particular, instead of the modulus non-linearity we opt to work with power coefficients and, therefore, use the squared modulus non-linearity. We quantify the robustness of scattering features using the word error rates of acoustic models trained on clean speech and evaluated using sets of utterances corrupted with different noise types. Our empirical results show that the second order scattering power spectrum coefficients capture invariants relevant for noise robustness and that this additional information improves generalization to unseen noise conditions (almost 20% relative error reduction on AURORA4). This finding can have important consequences on speech recognition systems that typically discard the second order information and keep only the first order features (known for emulating MFCC and FBANK values) when representing speech.

show abstract

Section: Network Architecturementioning

confidence: 99%

Section: Network Architecturementioning

confidence: 99%

See 1 more Smart Citation

Deep Scattering Power Spectrum Features for Robust Speech Recognition

et al. 2020

View full text Add to dashboard Cite

show abstract

“…WST has enjoyed significant success in various audio [27] and biomedical [30] signal classification tasks. WST demonstrated promising results on the TIMIT dataset for phonetic classification [31] and recognition [32].…”

Section: Introductionmentioning

confidence: 99%

Hybrid Network For End-To-End Text-Independent Speaker Identification

Ghezaiel

Brun

Lézoray

2021

2020 25th International Conference on Pattern Recognition (ICPR)

View full text Add to dashboard Cite

Deep learning has recently improved the performance of Speaker Identification (SI) systems. Promising results have been obtained with Convolutional Neural Networks (CNNs). This success is mostly driven by the advent of large datasets. However in the context of decentralized commercial applications, collection of large amount of training data is not always possible. In addition, robustness of a SI system is adversely effected by short utterances. Therefore, in this paper, we propose a novel textindependent speaker identification system able to identify speakers by learning from only few training short utterances examples. To achieve this, we combine a two-layer wavelet scattering network coupled with a CNN. The proposed architecture takes variable length speech segments. To evaluate the effectiveness of the proposed approach, Timit and Librispeech datasets are used in the experiments. Our experiments shows that our hybrid architecture provides satisfactory results under the constraints of short and limited number of utterances. These experiments also show that our hybrid architecture are competitive with the state of the art.

show abstract

“…The restriction to wavelets filters allows the deep scattering networks to have explicit and physics-related properties (frequency band, timescales of interest, amplitudes) that greatly simplifies the architecture design in contrast with classical deep convolutional neural network. Scattering networks have shown to perform high-quality classification of audio signals [20][21][22] and electrocardiograms 23 . A deep scattering network decomposes the signal's structure through a tree of wavelet convolutions, modulus operations, and average pooling, providing a stable representation at multiple time and frequency scales 20 .…”

mentioning

confidence: 99%

Clustering earthquake signals and background noises in continuous seismic data with unsupervised deep learning

et al. 2020

View full text Add to dashboard Cite

The continuously growing amount of seismic data collected worldwide is outpacing our abilities for analysis, since to date, such datasets have been analyzed in a human-expertintensive, supervised fashion. Moreover, analyses that are conducted can be strongly biased by the standard models employed by seismologists. In response to both of these challenges, we develop a new unsupervised machine learning framework for detecting and clustering seismic signals in continuous seismic records. Our approach combines a deep scattering network and a Gaussian mixture model to cluster seismic signal segments and detect novel structures. To illustrate the power of the framework, we analyze seismic data acquired during the June 2017 Nuugaatsiaq, Greenland landslide. We demonstrate the blind detection and recovery of the repeating precursory seismicity that was recorded before the main landslide rupture, which suggests that our approach could lead to more informative forecasting of the seismic activity in seismogenic areas.

show abstract

Deep Scattering Spectrum with deep neural networks

Cited by 31 publications

References 25 publications

Deep Scattering Power Spectrum Features for Robust Speech Recognition

Deep Scattering Power Spectrum Features for Robust Speech Recognition

Hybrid Network For End-To-End Text-Independent Speaker Identification

Clustering earthquake signals and background noises in continuous seismic data with unsupervised deep learning

Contact Info

Product

Resources

About