Improving Generalization of End-to-End ASR through Diversity and Independence Regularization

Ye-Eun Ko, Mun-Hak Lee, Donghyun Kim, Joon‐Hyuk Chang · Interspeech · 2025

Summary

A novel regularization technique applicable to various ASR models: diversity loss and independence loss and e-pendence loss are proposed, which improve the model generalization performance and robustness through extensive experiments.

Type: article
Published: 2025-08-17
Cited by: 1
References: 33
DOI: https://doi.org/10.21437/interspeech.2025-1309
OpenAlex: https://openalex.org/W4415433411
Semantic Scholar: https://api.semanticscholar.org/CorpusID:282345588

Keywords

Generalization, Regularization (linguistics), Independence (probability theory), Pattern recognition (psychology)

References

Librispeech: An ASR corpus based on public domain audio booksVassil Panayotov, Guoguo Chen, Daniel Povey — 2015
End-to-end Continuous Speech Recognition using Attention-based Recurrent NN: First ResultsJan Chorowski, Dzmitry Bahdanau, Kyunghyun Cho — 2014
Deep Speech: Scaling up end-to-end speech recognitionAwni Hannun, Carl Case, Jared Casper — 2014
Dropout: a simple way to prevent neural networks from overfittingNitish Srivastava, Geoffrey E. Hinton, Alex Krizhevsky — 2014
A tutorial on hidden Markov models and selected applications in speech recognitionL. R. Rabiner — 1989
Speech recognition with deep recurrent neural networksAlex Graves, Abdelrahman Mohamed, Geoffrey E. Hinton — 2013
Enhancing the TED-LIUM Corpus with Selected Data for Language Modeling and More TED TalksAnthony Rousseau, Paul Deléglise, Yannick Estève — 2014
The third ‘CHiME’ speech separation and recognition challenge: Dataset, task and baselinesJon Barker, Ricard Marxer, Emmanuel Vincent — 2015
A Neural TransducerNavdeep Jaitly, David Sussillo, Quoc V. Le — 2015
Diversity regularization in deep ensemblesChangjian Shui, Azadeh Sadat Mozafari, Jonathan Marek — 2018
fairseq: A Fast, Extensible Toolkit for Sequence ModelingMyle Ott, Sergey Edunov, Alexei Baevski — 2019
SpecAugment: A Simple Data Augmentation Method for Automatic Speech RecognitionDaniel Park, William Chan, Yu Zhang — 2019
Improving CTC Using Stimulated Learning for Sequence ModelingJahn Heymann, Khe Chai Sim, Bo Li — 2019
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate ShiftSergey Ioffe, Christian Szegedy — 2015
ESPnet: End-to-End Speech Processing ToolkitShinji Watanabe, Takaaki Hori, Shigeki Karita — 2018
Towards End-to-End Speech Recognition with Deep Convolutional Neural NetworksYing Zhang, Mohammad Pezeshki, Philémon Brakel — 2016
Self-Attention Transducers for End-to-End Speech RecognitionZhengkun Tian, Jiangyan Yi, Jianhua Tao — 2019
Character-Aware Attention-Based End-to-End Speech RecognitionZhong Meng, Yashesh Gaur, Jinyu Li — 2019
Understanding Contrastive Representation Learning through Alignment and Uniformity on the HypersphereTongzhou Wang, Phillip Isola — 2020
Efficient Diversity-Driven Ensemble for Deep Neural NetworksWentao Zhang, Jiawei Jiang, Yingxia Shao — 2020

Cited by

No citing papers recorded for this paper.

Related papers

No related papers recorded.