Speech acoustic modeling from raw multichannel waveforms

By Yedid Hoshen, Ron J. Weiss, Kevin Wilson. 2015

240 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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 Speech & Audio

In Signal Processing

Citations in the corpus

WaveNet: A Generative Model for Raw Audio (2016, methodology): “Traditionally, speech recognition research has largely focused on using log mel-ﬁlterbank energies or mel-frequency cepstral coefﬁcients (MFCCs), but has been moving to raw audio recently (Palaz et al., 2013; T ¨ uske et al., 2014; Hoshen et al., 2015; Sainath et al., 2015).”
Learning Filterbanks from Raw Speech for Phone Recognition (2017, background): “[6, 7] propose an architecture composed of a convolutional layer followed by max-pooling and a nonlinearity, so that gammatone filterbanks correspond to a particular configuration of the network.”
Fully Convolutional Speech Recognition (2018, background): “Several proposals to learn the front-end of speech recognition systems have been made [16, 17, 14, 18].”
Recent advances in convolutional neural networks (2015, background): “CNNs, and to learn filters to process the raw waveform jointly with the rest of the network [291, 292].”
Dnsmos: A Non-Intrusive Perceptual Objective Speech Quality Metric to Evaluate Noise Suppressors (2020, background): “Recently, researchers have seen success in learning features within the model for tasks such as SE [17], speech, and music synthesis [18] and to learn acoustic models [19].”
Dnsmos P.835: A Non-Intrusive Perceptual Objective Speech Quality Metric to Evaluate Noise Suppressors (2021, background): “Recently, researchers have seen success in learning features within the model for tasks such as SE [25], speech, and music synthesis [26] and to learn acoustic models [27].”
Automatic Sleep Stage Scoring with Single-Channel EEG Using Convolutional Neural Networks (2016, methodology): “Again, there are characteristic examples from speech and acoustics in [6, 19, 27, 20, 10].”
Multichannel Signal Processing With Deep Neural Networks for Automatic Speech Recognition (2017, methodology): “In order to produce an output that is invariant to perceptually and semantically identical sounds appearing at different time shifts we pool the outputs in time after filtering [14], [15], in an operation that has an effect similar to discarding the phase in the short-time Fourier transform.”
Neural network based spectral mask estimation for acoustic beamforming (2016, background): “And despite recent attempts to take advantage of multi-channel data within a (convolutional) DNN or even training a network directly on multi-channel waveforms, model-based beamforming still proved to be superior [1] [2].”
Deep Learning for Environmentally Robust Speech Recognition (2017, methodology): “ers and post-ﬁlters for speech enhancement, joint front- and back-end multi-channel ASR systems have recently attracted considerable attention with a goal of decreasing the WER directly [131], [132], [134]. In [130], the individual features extracted from each microphone channel are concatenated as ”
Interpretable Convolutional Filters with SincNet (2018, background): “This approach has shown promise in speech [25, 26, 27, 28, 29], including emotion tasks [30], speaker recognition [35], spoofing detection [34], and speech synthesis [31, 32].”
Deep Learning for Audio Signal Processing (2019, methodology): “[18] and [19] use a full-resolution magnitude spectrum, [20]–[23] directly use a raw waveform representation of the audio signals as inputs and learn datadriven filters jointly with the rest of the network for the target tasks.”
New Era for Robust Speech Recognition (2017, methodology): “It can be applied to distributed microphone arrays [22], but this chapter describes three similar systems for performing unsupervised spatial clustering and beamforming with compact microphone arrays [29, 37, 49].”
Acoustic Modelling from the Signal Domain Using CNNs (2016, methodology): “The filters learned by our network are similar to those observed in other raw waveform works in the literature [5, 7, 22].”
Learning the speech front-end with raw waveform CLDNNs (2015, methodology): “Past work has achieved phase invariance using convolutional layers which pool in time [4, 5, 7] or DNN layers with large, potentially overcomplete, hidden units [6], which can capture the same filter shape at a variety of phases.”
Neural Network Adaptive Beamforming for Robust Multichannel Speech Recognition (2016, background): “However, one shortcoming of previous work [1, 2, 3] is that the filters learned during training are fixed for decoding, potentially limiting the ability of these models to adapt to previously unseen or changing conditions.”
Speaker Recognition from Raw Waveform with SincNet (2018, background): “This approach has shown promise in speech [22–26], including emotion tasks [27], speaker recognition [32], spoofing detection [31], and speech synthesis [28, 29].”