CTC in the Context of Generalized Full-Sum HMM Training

By Albert Zeyer, Eugen Beck, Ralf Schlüter, Hermann Ney. 2017

43 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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. Jinyu Li
2. Dong Yu
3. Shinji Watanabe
4. Ralf Schlüter
5. Hermann Ney
6. Tara N. Sainath
7. Li Deng
8. Takaaki Hori

In Computer Vision and Pattern Recognition

1. Jinyu Li
2. Hermann Ney
3. Ralf Schlüter
4. Albert Zeyer
5. Kazuki Irie
6. Li Deng
7. Takaaki Hori
8. Sanjeev Khudanpur

Citations in the corpus

• Unsupervised Speech Recognition (2021, neutral): “The HMM associated with CTC has one state for each letter and a special state for the blank symbol (Zeyer et al., 2017).”
• Improved training of end-to-end attention models for speech recognition (2018, background): “There was certain work to remove the GMM dependency in the pretraining [5], or to be able to train without an existing alignment [6–8], or to avoid the lexicon [9], which simplifies the pretraining procedure but still is not end-to-end.”
• RETURNN as a Generic Flexible Neural Toolkit with Application to Translation and Speech Recognition (2018, background): “5 tional LSTM with 6 layers (Zeyer et al., 2017b), and a generalized full-sum sequence trained hybrid deep bidirectional LSTM with 5 layers (Zeyer et al., 2017a).”
• Recent Advances in End-to-End Automatic Speech Recognition (2021, methodology): “CTC is the first E2E technology widely used in ASR [7, 80, 153, 188, 259, 324, 351].”
• End-to-end Speech Recognition Using Lattice-free MMI (2018, methodology): “a character, in character-based CTC) has a special 2-state HMM topology as shown in Figure 1a [24].”
• Knowledge Distillation for Sequence Model (2018, background): “Several works comparing traditional systems with CTC based systems also show that the quality of σCTC(k, t) is the bottleneck of the convergence speed and performance [15].”
• End-to-End Speech Recognition: A Survey (2023, background): “[12], [13], [33], [211] handle the alignment problem in a way that can be interpreted to be similar to HMMs with a specific model topology, including marginalization over alignments [90], [248], [249].”