List your company
ResearchFormerly WizWand

Automatic Speech Recognition on Hub5 2000 (SWB)

5.8WER2016 Microsoft system
5.6726.5367.48.264Apr 27, 2016Dec 31, 2016Sep 6, 2017May 12, 2018Jan 16, 2019Sep 21, 2019May 27, 2020
Updated 8mo ago

Evaluation results

MethodLinks
2016 Microsoft systemLanguage Model=Neural net LM2016.105.8
Human transcription2016.105.9
n-gram + model M + NNLMLanguage Model=n-gram + model M + NNLM2016.046.6
Saon et al. systemLanguage Model=Neural net LM2016.106.6
n-gram + NNLMLanguage Model=n-gram + NNLM2016.046.8
n-gram + model MLanguage Model=n-gram + model M2016.047
CATUnit=mono-phone, LM=4-gram, RNN-LM rescoring=true2020.057.3
2016 Microsoft systemLanguage Model=N-gram LM2016.107.4
LF-MMIUnit=bi-phone, LM=4-gram, RNN-LM rescoring=true2020.057.5
85K vocab n-gram baselineLanguage Model=85K vocab, 36M n-grams2016.047.6
Saon et al. systemLanguage Model=N-gram LM2016.107.6
E2E-LF-MMIUnit=bi-phone, LM=4-gram, RNN-LM rescoring=true2020.057.6
CATUnit=mono-phone, LM=4-gram, RNN-LM rescoring=false2020.057.9
RNN-TUnit=char, LM=4-gram, RNN-LM rescoring=false2020.058.1
AttentionUnit=char, LM=No LM, RNN-LM rescoring=false2020.058.3
LF-MMIUnit=bi-phone, LM=4-gram, RNN-LM rescoring=false2020.058.4
Povey et al. LSTMLanguage Model=N-gram LM2016.108.5
30K vocab n-gram baselineLanguage Model=30K vocab, 4M n-grams2016.048.6
E2E-LF-MMIUnit=bi-phone, LM=4-gram, RNN-LM rescoring=false2020.058.6
E2E-LF-MMIUnit=mono-phone, LM=4-gram, RNN-LM rescoring=false2020.058.9
Saon et al. LSTMLanguage Model=N-gram LM2016.109