Cross-Domain and Cross-Language Portability of Acoustic Features Estimated by Multilayer Perceptrons

By Andreas Stolcke, Frantisek Grézl, Mei-Yuh Hwang, X. Lei, Nelson Morgan, D. Vergyri. 2006

137 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

1. Dong Yu
2. Li Deng
3. Jinyu Li
4. Yoshua Bengio
5. Shinji Watanabe
6. James R. Glass
7. Quoc V. Le
8. Yifan Gong

In Speech & Audio

1. Jinyu Li
2. Dong Yu
3. Shinji Watanabe
4. Li Deng
5. S. Renals
6. James R. Glass
7. Yifan Gong
8. Pawel Swietojanski

In Signal Processing

1. Dong Yu
2. Jinyu Li
3. Li Deng
4. Yifan Gong
5. S. Renals
6. Shinji Watanabe
7. Florian Metze
8. Yoshua Bengio

Citations in the corpus

• Unsupervised Pretraining Transfers Well Across Languages (2020, background): “A common way to improving speech recognition in low-resource languages is to train multilingual speech recognition with shared components [8, 9, 2].”
• Impact of deep MLP architecture on different acoustic modeling techniques for under-resourced speech recognition (2013, unknown): “In this paper, we investigate the performance of posterior based approaches in the context of under-resourced speech recognition when a standard three-layer MLP is replaced by a deeper fivelayer MLP.”
• Cross-language knowledge transfer using multilingual deep neural network with shared hidden layers (2013, background): “The value of cross-lingual transfer has also been investigated in multilayer perceptrons (MLPs) under the tandem framework [20][21].”
• Multi-level adaptive networks in tandem and hybrid ASR systems (2013, methodology): “The hidden layers use logistic sigmoid nonlinearities and a softmax is used for the output layer to provide posterior probability estimates for each output class.”
• Sequence-Based Multi-Lingual Low Resource Speech Recognition (2018, background): “With the on-set of deep learning the focus of the models shifted to learning features across languages which can be mapped to the same space [3, 11].”
• Massively Multilingual Adversarial Speech Recognition (2019, background): “…an acoustic model trained on one language to effectively recognize the sounds of other languages (Schultz and Waibel, 2001; Le and Besacier, 2005; Stolcke et al., 2006; T´oth et al., 2008; Plahl et al., 2011; Thomas et al., 2012; Imseng et al., 2014; Do et al., 2014; Heigold et al., 2013;…”
• Automatic speech recognition for under-resourced languages: A survey (2014, background): “Several studies showed that features extracted from an MLP which was trained with one or multiple languages can be applied to other languages (Stolcke et al., 2006; Toth et al., 2008; Plahl et al., 2011).”
• Current trends in multilingual speech processing (2011, methodology): “In Stolcke et al (2006), it was shown that using MLP-based features extracted from English-trained MLP could improve 16http://www.clsp.jhu.edu/ws2006/groups/afsr Mandarin and Arabic ASR performance over the spectral feature baseline system.”
• Multilingual acoustic models using distributed deep neural networks (2013, methodology): “Overfitting remains an issue, unless the hypothesis space is reduced by using regularization (similar to model adaptation [13, 14]) or keeping the bottom hidden layers fixed during training (see [15] and feature learning in Section 2.”
• Deep neural network features and semi-supervised training for low resource speech recognition (2013, background): “The key element in this data-driven approach is a multi-layer perceptron (MLP) trained on large amounts of task independent data, i.e. multilingual data or data from the same language but collected under different settings [6, 7].”
• Investigation on cross- and multilingual MLP features under matched and mismatched acoustical conditions (2013, background): “Since it was first observed in [5] that the concatenated cepstral and MLP based posterior features trained on English data significantly improved the MFCC based systems in entirely different languages, Arabic or Mandarin, many studies investigated the crosslingual portability of NNs.”
• Automatic speech recognition and speech variability: A review (2007, background): “A important study of the robustness of similarly obtained MLP-based acoustic features to domains and languages is also reported in Stolcke et al. (2006) .”
• Multilingual training of deep neural networks (2013, background): “In such approaches, networks trained on a source language are used to provide features for a target language [17, 18].”
• Unsupervised cross-lingual knowledge transfer in DNN-based LVCSR (2012, background): “Examples include: the direct use of posterior features obtained from a source language network [17]; the use of cross-lingual bottleneck features [3]; training/initialising a neural network using transcribed source language acoustics, then retraining the”