Recurrent deep neural networks for robust speech recognition

By Chao Weng, Dong Yu, Shinji Watanabe, Biing-Hwang Fred Juang. 2014

139 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

1. Shinji Watanabe
2. Jinyu Li
3. Dong Yu
4. John R. Hershey
5. James R. Glass
6. Bo Li
7. Tara N. Sainath
8. Yoshua Bengio

In Signal Processing

1. Shinji Watanabe
2. Jinyu Li
3. Deliang Wang
4. John R. Hershey
5. Dong Yu
6. Jonathan Le Roux
7. Zhuo Chen
8. Bo Li

Citations in the corpus

• Power of Deep Learning for Channel Estimation and Signal Detection in OFDM Systems (2017, background): “Deep learning has been successfully applied in a wide range of areas with significant performance improvement, including computer vision [6], natural language processing [7], speech recognition [8], and so on.”
• Unsupervised neural network based feature extraction using weak top-down constraints (2015, methodology): “To use it as such, the encoding obtained from one of its middle layers is finally taken as the feature representation of new input speech, as illustrated in the right-most block of Figure 1.”
• Towards Deep Developmental Learning (2016, background): “development of writing capabilities and other language related processes [7]. learning techniques have shown their efficiency at dealing with many of these language related issues [83], [84].”
• Building DNN acoustic models for large vocabulary speech recognition (2014, background): “Modern recurrent network approaches to acoustic modeling have shown some initial success on large vocabulary tasks [32], and tasks where limited training data is available [33], [34], [35], [36].”
• Neural Network Adaptive Beamforming for Robust Multichannel Speech Recognition (2016, background): “While automatic speech recognition (ASR) performance has improved dramatically in recent years, particularly with the advent of deep learning [5], performance in realistic noisy and far-field scenarios is still far-behind clean speech conditions [6, 7, 8] To improve robustness, microphone arrays are”
• Speech Enhancement with LSTM Recurrent Neural Networks and its Application to Noise-Robust ASR (2015, methodology): “Compared to a similar system that uses BF and DNN-based masking as a front-end for a DNN acoustic model [7], we obtain a 20 % relative improvement (from 18.0 %).”
• The third ‘CHiME’ speech separation and recognition challenge: Dataset, task and baselines (2015, unknown): “It is based on the Kaldi recipe for Track 2 of the 2nd CHiME Challenge [16].”
• Improving Robustness of Deep Neural Network Acoustic Models via Speech Separation and Joint Adaptive Training (2015, methodology): “Note that a similar strategy was proposed in [3].”
• Robust speech recognition using long short-term memory recurrent neural networks for hybrid acoustic modelling (2014, background): “Such systems proved to be very robust in adverse conditions due to their increased modelling power [3, 4].”
• Deep Neural Networks for Single-Channel Multi-Talker Speech Recognition (2015, unknown): “Since DNNs and RNNs can extract more in­variant features than the raw acoustic feature at the output and hidden layers [17], they can also be used to generate the so called Tandem [24], [25] feature if it is extracted from the output layer, or bottleneck feature [26], [27] if it is extracted…”
• Discriminatively trained recurrent neural networks for single-channel speech separation (2014, background): “recurrent neural networks have seen a resurgence in popularity for speech and music processing tasks [9], [11]–[15].”
• Deep Reinforcement Learning Based Resource Allocation for V2V Communications (2018, background): “Recently, deep learning has made great stride in speech recognition [17], image recognition [16], and wireless communications [18].”
• Transferring knowledge from a RNN to a DNN (2015, methodology): “Recently, Recurrent Neural Networks (RNNs) have demonstrated even more potential over its DNN counterparts [6, 7, 8].”
• Speech enhancement and recognition using multi-task learning of long short-term memory recurrent neural networks (2015, background): “Also note that the best reported WER for the same task in [17] and [18] are 26.”
• Uncertainty propagation through deep neural networks (2015, methodology): “The baseline ASR system has been trained as follows [16, 17].”
• Maximum a posteriori Based Decoding for CTC Acoustic Models (2016, background): “Encouraged by the strong performance of DNNs, various extensions of the network architecture have been proposed, such as convolutional neural networks (CNNs) [6], sigmoid-unit-type recurrent neural networks (RNNs) [7, 8, 9, 10], and long short-term memory neural networks [11, 12, 13, 14, 15].”
• Deep neural network based spectral feature mapping for robust speech recognition (2015, methodology): “Recurrent neural networks (RNNs) have been applied in a tandem system [8] or in a hybrid system with deep structure [9] to model temporal dependency of speech and noise for robust ASR.”