A segmental CRF approach to large vocabulary continuous speech recognition

By Geoffry Zweig, Patrick Nguyen. 2009

137 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. Li Deng
2. Dong Yu
3. Jinyu Li
4. Yoshua Bengio
5. Jianfeng Gao
6. James R. Glass
7. Karen Livescu
8. Mohit Bansal

In Signal Processing

1. Dong Yu
2. Li Deng
3. Jinyu Li
4. Ossama Abdel-Hamid
5. Hui Jiang
6. George E. Dahl
7. Yifan Gong
8. Abdel-rahman Mohamed

Citations in the corpus

• Deep Learning: Methods and Applications (2014, background): “Many of the discriminative techniques for supervised learning in signal and information processing are shallow architectures such as HMMs [52, 127, 147, 186, 188, 290, 394, 418] and conditional random fields (CRFs) [151, 155, 281, 400, 429, 446].”
• Discriminative acoustic word embeddings: Tecurrent neural network-based approaches (2016, methodology): “Maas et al. trained CNNs to predict (continuous-valued) embeddings of the word labels, and used the resulting embeddings to define feature functions in a segmental conditional random field [18] rescoring system.”
• Context-Dependent Pre-Trained Deep Neural Networks for Large-Vocabulary Speech Recognition (2012, methodology): “Since we did not personally obtain the result from [70], there may be other differences between our setup and the one used in [70] in addition to the larger training set.”
• Exemplar-Based Sparse Representation Features: From TIMIT to LVCSR (2011, unknown): “For example, CRFs lend itself to better optimization techniques and can allow for multiple feature streams [ 49 ].”
• Learning Word Embeddings: Unsupervised Methods for Fixed-size Representations of Variable-length Speech Segments (2018, unknown): “The framework of segmental speech recognition [12] involves an explicit segmentation of speech utterances and a fixed-size representation for each acoustic segment.”
• Deep segmental neural networks for speech recognition (2013, methodology): “In contrast, the LM for a SCRF is defined using the transi- tional features and their weights between two states.”
• Integrating Deep Neural Networks into Structural Classification Approach based on Weighted Finite-State Transducers (2012, background): “For example, [5] provides a flexible framework that can integrate several detectors to realize accurate recognition.”
• Large vocabulary continuous speech recognition with context-dependent DBN-HMMS (2011, methodology): “First, in order to compare our results with [13], we would need to compute sentence accuracy.”
• Word embeddings for speech recognition (2014, background): “A recent line of work includes segmental conditional random fields [18] with word template features [12, 5], for example.”