Acoustic modeling with deep neural networks using raw time signal for LVCSR

By Zolt'an Tuske, Pavel Golik, Ralf Schlüter, Hermann Ney. 2014

208 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Oceanography, 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 Oceanography

In Signal Processing

Citations in the corpus

Analysis of CNN-based speech recognition system using raw speech as input (2015, methodology): “Using temporal raw speech directly as input has been proposed in the context of DNNs [14] and in the context of end-to-end sequence-discriminative training of CNNs [15].”
Learning Multiscale Features Directly from Waveforms (2016, background): “While many studies have compared convolution on raw waveforms to features such as spectrograms, MFCCs, and melscale filterbanks, they often compare the two at a similar strides and window sizes [5, 6].”
On internal language representations in deep learning: an analysis of machine translation and speech recognition (2018, background): “Others have investigated learning from the raw waveform and also found benefit in combining multiple representations as input to the neural network [325].”
Multichannel Signal Processing With Deep Neural Networks for Automatic Speech Recognition (2017, background): “Previous work has demonstrated the possibility of training deep networks directly on raw, single channel, time domain waveform samples [10], [11], [12], [13], [14], [15].”
End-to-End Speech Recognition: A Survey (2023, methodology): “integrating speech signal preprocessing and feature extraction into acoustic modeling [10], [11].”
Deep Learning for Audio Signal Processing (2019, background): “hidden units [22], which are able to capture the same filter shape at a variety of phases.”
Speech acoustic modeling from raw multichannel waveforms (2015, background): “This result extends the results in [2, 3] to multicondition training with noise and reverberation.”
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].”
Raw waveform-based speech enhancement by fully convolutional networks (2017, background): “In the field of ASR, several studies have shown that deeplearning-based models with raw waveform inputs can achieve higher accuracy than those with hand-crafted features (e.g., MFCC) [22-27].”
A deep auto-encoder based low-dimensional feature extraction from FFT spectral envelopes for statistical parametric speech synthesis (2016, methodology): “[8, 9]), and some papers have achieved a relative reduction in word error rates using a combination of features derived from raw waveforms and log-mel features [10].”
Acoustic Modelling from the Signal Domain Using CNNs (2016, methodology): “[4] proposed to use the raw time signal directly as input to a DNN.”
Very deep convolutional neural networks for raw waveforms (2016, background): “Recent works have applied CNNs to audio tasks such as environmental sound recognition and speech recognition and found that CNNs perform well with just the raw waveforms [9, 4, 10].”
Speaker Recognition from Raw Waveform with SincNet (2018, background): “To mitigate this drawback, some recent works have proposed directly feeding the network with spectrogram bins [18–20] or even with raw waveforms [21–30].”
Learning Speaker Representations with Mutual Information (2018, unknown): “Similarly to [25–29], we propose to directly process raw waveforms rather than using standard MFCC, or FBANK features.”
Convolutional Neural Networks-based continuous speech recognition using raw speech signal (2015, methodology): “More recently, ANNs with deep learning architectures, more precisely, deep belief network or deep neural networks (DNNs) [7, 8], which can yield better system than a single hidden layer MLP have been proposed to address various aspects of acoustic modeling.”