Very Deep Convolutional Neural Networks for Noise Robust Speech Recognition

By Yanmin Qian, Mengxiao Bi, Tian Tan, Kai Yu. 2016

332 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Model Architectures, 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

1. Dong Yu
2. Jinyu Li
3. Li Deng
4. Jasha Droppo
5. Yu Zhang
6. Yoshua Bengio
7. Shinji Watanabe
8. M. Seltzer

In Model Architectures

1. Jasha Droppo
2. Yu Zhang
3. William Chan
4. Andreas Stolcke
5. Dong Yu
6. Geoffry Zweig
7. M. Seltzer
8. Xiangang Li

In Speech & Audio

1. Dong Yu
2. Jinyu Li
3. Jasha Droppo
4. Yu Zhang
5. Li Deng
6. Andreas Stolcke
7. Xiangang Li
8. M. Seltzer

In Signal Processing

1. Dong Yu
2. Jinyu Li
3. Li Deng
4. Jasha Droppo
5. Xiangang Li
6. Yu Zhang
7. M. Seltzer
8. Shinji Watanabe

Citations in the corpus

• Improving Transformer-based Speech Recognition Using Unsupervised Pre-training (2019, background): “Apart from speaking style, other speech variabilities like speaking environment, speech scenarios and regional dialect [26, 30, 31] also have an impact on the performance of speech recognition system.”
• Achieving Human Parity in Conversational Speech Recognition (2016, background): “Surprisingly, this is the case for both acoustic modeling [28, 29, 30, 31, 32, 33] and language modeling [34, 35, 36, 37].”
• Very deep convolutional networks for end-to-end speech recognition (2016, background): “Recently, very deep CNNs architectures [15] have also been shown to be successful in ASR [16, 17], using more non-linearities, but fewer parameters.”
• The microsoft 2016 conversational speech recognition system (2016, methodology): “tions of CNNs to speech recognition [41, 42], our networks are much deeper, and use linear bypass connections across convolutional layers. They are similar in spirit to those studied more recently by [11, 10, 31, 12, 13]. We improve on these architectures with the LACE model [19], which iteratively ”
• Deep and Wide Neural Networks on Multiple sets of Temporal data with Correlation (2018, background): “Many of the early concepts and models of neural networks [1,2,3,4] produced new vitality after combining them with Deep Neural Networks, they have made a strong push in acoustic modeling [5,6], language modeling [7], machine vision and so on.”
• The Microsoft 2017 Conversational Speech Recognition System (2017, unknown): “The combination of deep networks and efficient training methods with older neural modeling concepts [2, 3, 4, 5, 6, 7, 8] have produced steady advances in both acoustic modeling [9, 10, 11, 12, 13, 14, 15] and language modeling [16, 17, 18, 19].”
• Comparing Human and Machine Errors in Conversational Speech Transcription (2017, methodology): “Using this methodology, and incorporating state-of-the-art convolutional and recurrent network architectures for both acoustic modeling [10, 11, 8, 12, 13, 14, 15] and language modeling [16, 17, 18, 19] with extensive use of model combination, we obtained a machine error rate that was very slightly ”
• Deep Learning for Environmentally Robust Speech Recognition (2017, methodology): “Compared with DBMs, the CNNs have the advantages (Qian et al. 2016): (1) they are well suited to model the local correlations in both time and frequency in speech spectrogram and (2) translational invariance, such as the frequency shift due to speaker or speaking style variations, can be more…”
• Hierarchical Bayesian combination of plug-in maximum a posteriori decoders in deep neural networks-based speech recognition and speaker adaptation (2017, background): “filters, pooling, and input feature maps in a convolutional architecture requires an expensive tuning phase, as discussed in (Qian et al., 2016).”
• Multi-head CNN-RNN for multi-time series anomaly detection: An industrial case study (2019, methodology): “The CNN-RNN architecture has achieved promising results in various domains such as speech recognition [30], gesture recognition [31], weather recognition [59], or emotion detection [60].”
• Toward Human Parity in Conversational Speech Recognition (2017, background): “This is the case for both acoustic modeling [28]–[33] and language modeling [34]–[37].”
• A novel pyramidal-FSMN architecture with lattice-free MMI for speech recognition (2018, unknown): “ons [4] and utilization of sequential information. Convolutional neural network (CNN), in which local connectivity, weight sharing, and pooling techniques are applied, also outperforms previous works [8, 9]. However, the training of RNN relies on backpropagation through time (BPTT) [10], which may b”
• Deep Feed-Forward Sequential Memory Networks for Speech Synthesis (2018, background): “to gradient vanishing problem than doing back-propagation through time (BPTT) with recurrent networks. 2.3. Deep-FSMN Inspired by successful applications of very deep networks in many different fields [18, 19, 20, 21, 22, 23], one potential improvement over cFSMN is to make the network deeper. We giv”