Learning filter banks within a deep neural network framework

By Tara N. Sainath, Brian Kingsbury, Abdel-rahman Mohamed, Bhuvana Ramabhadran. 2013

175 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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 Signal Processing

Citations in the corpus

Deep Learning: Methods and Applications (2014, methodology): “Substantial speech recognition error reduction is reported in [307].”
Analysis of CNN-based speech recognition system using raw speech as input (2015, methodology): “CNN-based system taking power spectrum as input has been proposed in [13].”
Acoustic modeling with deep neural networks using raw time signal for LVCSR (2014, background): “[6], or even the raw audio samples of the waveform.”
An Overview of Noise-Robust Automatic Speech Recognition (2014, methodology): “And in [292], the filter bank is replaced with a filter bank layer that is learned jointly with the rest of DNN by taking the linear frequency power spectrum as the input.”
Filterbank learning for deep neural network based polyphonic sound event detection (2016, methodology): “This led to a trend of using raw data (pixel values for an image [24] or magnitude/power spectrum for audio [25], [26], [27] or even raw audio [28]) as input and deep learning methods as the classifier in machine learning tasks.”
Speaker Recognition from Raw Waveform with SincNet (2018, methodology): “Similar to SincNet, some previous works have proposed to add constraints on the CNN filters, for instance forcing them to work on specific bands [33, 34].”
A survey of deep neural network architectures and their applications (2017, methodology): “Moreover, the DNNs have been employed for tracking dialog state [53], transferring cros s-language knowledge [71], learning the filter banks [135], designing an automatic feature extraction systems from aud io [51], and speaker adaptive training (SAT) for acoustic modeling [114].”
Acoustic Modelling from the Signal Domain Using CNNs (2016, background): “With DNNs, mel-like filterbanks can be learnt from the power spectrum [3].”
Speech acoustic modeling from raw multichannel waveforms (2015, methodology): “[11] improve performance by training a model on linear frequency FFf magnitude features to learn filterbank weights automatically rather than relying on the mel scale [12].”
Discriminative template learning in group-convolutional networks for invariant speech representations (2015, methodology): “Learning the filters jointly with the network parameters has been explored as an alternative to the base Mel-filter responses [17].”
Deep Scattering Spectrum with deep neural networks (2014, background): “For example, [1] looked at starting with a raw acoustic representation of the signal (i.”
Interpretable Convolutional Filters with SincNet (2018, methodology): “Similar to SincNet, some previous works have proposed to add constraints on the CNN filters, for instance forcing them to work on specific bands [46, 47].”