GPU-ACCELERATED FORWARD-BACKWARD ALGORITHM WITH APPLICATION TO LATTICE-FREE MMI

被引：3

作者：

Ondel, Lucas ^{[1
]}

Lam-Yee-Mui, Lea-Marie ^{[1
,3
]}

Kocour, Martin ^{[2
]}

Corro, Caio Filippo ^{[1
]}

Burget, Lukas ^{[2
]}

机构：

[1] Univ Paris Saclay, LISN, CNRS, Orsay, France

[2] Brno Univ Technol, Fac Informat Technol, Brno, Czech Republic

[3] Vocapia Res, Orsay, France

来源：

2022 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP) | 2022年

关键词：

Lattice-Free MMI; end-to-end ASR; Julia language; forward-backward;

D O I：

10.1109/ICASSP43922.2022.9746824

中图分类号：

O42 [声学];

学科分类号：

070206 ; 082403 ;

摘要：

We propose to express the forward-backward algorithm in terms of operations between sparse matrices in a specific semiring. This new perspective naturally leads to a GPU-friendly algorithm which is easy to implement in Julia or any programming languages with native support of semiring algebra. We use this new implementation to train a TDNN with the LF-MMI objective function and we compare the training time of our system with PyChain-a recently introduced C++/CUDA implementation of the LF-MMI loss. Our implementation is about two times faster while not having to use any approximation such as the "leaky-HMM".

引用

页码：8417 / 8421

页数：5

共 50 条

[41] A FORWARD-BACKWARD ALGORITHM FOR REWEIGHTED PROCEDURES: APPLICATION TO RADIO-ASTRONOMICAL IMAGING
Repetti, Audrey
Wiaux, Yves
2020 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING, 2020, : 1434 - 1438
[42] SEMI-SUPERVISED TRAINING OF ACOUSTIC MODELS USING LATTICE-FREE MMI
Manohar, Vimal
Hadian, Hossein
Povey, Daniel
Khudanpur, Sanjeev
2018 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP), 2018, : 4844 - 4848
[43] A GPU-Accelerated Density-Based Clustering Algorithm
Loh, Woong-Kee
Kim, Young-Kuk
2014 IEEE FOURTH INTERNATIONAL CONFERENCE ON BIG DATA AND CLOUD COMPUTING (BDCLOUD), 2014, : 775 - 776
[44] GPU-Accelerated Algorithm for Online Probabilistic Power Flow
Zhou, Gan
Bo, Rui
Chien, Lungsheng
Zhang, Xu
Yang, Shengchun
Su, Dawei
IEEE TRANSACTIONS ON POWER SYSTEMS, 2018, 33 (01) : 1132 - 1135
[45] Autonomous sparse Markowitz portfolio based on two-stage accelerated forward-backward algorithm
Lin, Yizun
Wang, Linhui
Lai, Zhao-Rong
EXPERT SYSTEMS WITH APPLICATIONS, 2025, 271
[46] Accelerated forward-backward and Douglas-Rachford splitting dynamics
Ozaslan, Ibrahim K.
Jovanovic, Mihailo R.
AUTOMATICA, 2025, 175
[47] LATTICE-FREE MMI ADAPTATION OF SELF-SUPERVISED PRETRAINED ACOUSTIC MODELS
Vyas, ApoorV
Madikeri, Srikanth
Bourlard, Herve
2021 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP 2021), 2021, : 6219 - 6223
[48] Domain adaptation of lattice-free MMI based TDNN models for speech recognition
Long Y.
Li Y.
Ye H.
Mao H.
International Journal of Speech Technology, 2017, 20 (1) : 171 - 178
[49] Algorithm of GPU-accelerated color transfer between images
Li, Jian-Ming
Wan, Dan-Ling
He, Rong-Sheng
Qian, Kun-Ming
Dalian Ligong Daxue Xuebao/Journal of Dalian University of Technology, 2008, 48 (02): : 287 - 291
[50] Swift Unfolding of Communities: GPU-Accelerated Louvain Algorithm
Wang, Zhibin
Lin, Xi
Li, Xue
Wang, Pinhuan
Meng, Ziheng
Liu, Hang
Tian, Chen
Zhong, Sheng
PROCEEDINGS OF THE 2025 THE 30TH ACM SIGPLAN ANNUAL SYMPOSIUM ON PRINCIPLES AND PRACTICE OF PARALLEL PROGRAMMING, PPOPP 2025, 2025, : 441 - 454

← 1 2 3 4 5 →