BAYES RISK CTC: CONTROLLABLE CTC ALIGNMENT IN SEQUENCE-TO-SEQUENCE TASKS

Authors: Jinchuan Tian, Brian Yan, Jianwei Yu, CHAO WENG, Dong Yu, Shinji Watanabe

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimentally, the proposed BRCTC, along with a trimming approach, enables us to reduce the inference cost of offline models by up to 47% without performance degradation; BRCTC also cuts down the overall latency of online systems to an unseen level
Researcher Affiliation Collaboration Jinchuan Tian, Jianwei Yu , Chao Weng, Dong Yu Tencent AI LAB {tyriontian, tomasyu, cweng, dyu}@tencent.com Brian Yan & Shinji Watanabe Language Technologies Institute, Carnegie Mellon University, Pittsburgh, PA 15213, USA {byan, swatanab}@andrew.cmu.edu
Pseudocode No No pseudocode or clearly labeled algorithm block was found.
Open Source Code Yes Code release: https://github.com/espnet/espnet.
Open Datasets Yes Datasets: Experiments are mainly conducted for ASR, but MT and ST are also included. For ASR, the experiments are on Aishell-1 (Bu et al., 2017), Aishell-2 (Du et al., 2018), Wenetspeech (Zhang et al., 2022) and Librispeech (Panayotov et al., 2015).
Dataset Splits Yes Aishell-1 dev / test... Wenetspeech dev / meeting / net
Hardware Specification Yes All experiments are conducted on Nvidia V100 GPUs. All the inference jobs are conducted on Intel(R) Xeon(R) Platinum 8255C CPU (2.5GHz).
Software Dependencies No No specific version numbers for ancillary software dependencies were provided.
Experiment Setup Yes Implementation details are in appendix D and E for reproducibility. Details for the optimization and the setting for BRCTC are shown in table 7.