List your company
ResearchFormerly WizWand

Automatic Speech Recognition on 80-hour WSJ (dev93)

5.7WERVGG-BLSTM
5.1249.01212.916.788Nov 11, 2020Sep 19, 2021Jul 29, 2022Jun 7, 2023Apr 14, 2024Feb 21, 2025Dec 31, 2025
Updated 8mo ago

Evaluation results

MethodLinks
VGG-BLSTMTraining Criterion=CTC-CRF2020.115.7
ESPRESSO2020.115.9
SS-LF-MMI2020.116
BLSTMTraining Criterion=CTC-CRF2020.116.23
TDNN-DTraining Criterion=CTC-CRF2020.116.24
FC-SR2020.116.8
Masked-VPCLanguage Model=4-gram word language model, Fine-tuning Protocol=CTC2025.126.8
HuBERT ObjLanguage Model=4-gram word language model, Fine-tuning Protocol=CTC2025.127.3
Masked-NCELanguage Model=4-gram word language model, Fine-tuning Protocol=CTC2025.127.8
BLSTMTraining Criterion=CTC2020.118.57
EE-Policy-CTC2020.119.21
Masked-VPCLanguage Model=None, Fine-tuning Protocol=CTC2025.1214.1
Masked-NCELanguage Model=None, Fine-tuning Protocol=CTC2025.1214.7
HuBERT ObjLanguage Model=None, Fine-tuning Protocol=CTC2025.1215.5
wav2letter++Language Model=None, Fine-tuning Protocol=CTC2025.1219.5
12L TransformerLanguage Model=None, Fine-tuning Protocol=CTC2025.1220.1