List your company
ResearchFormerly WizWand

Automatic Speech Recognition on Tuda-DE (dev)

11.1WERIMS-Speech
9.731618.968328.20537.4417Jul 26, 2018Nov 23, 2018Mar 23, 2019Jul 22, 2019Nov 19, 2019Mar 18, 2020Jul 17, 2020
Updated 8mo ago

Evaluation results

MethodLinks
IMS-SpeechLanguage=German2019.0811.1
TransformerTraining Datasets=TD, SW*, LV*, Training data hours (h)=1460, LM=RNNLM (4 x 1024), Alignment=CTC-segmentation2020.0712.3
TDNN-HMMTraining Datasets=TD, SW, MA, Training data hours (h)=412, LM=LSTM (2 x 1024)2020.0713.1
State of the artLanguage=German2019.0813.1
TDNN-HMMTraining Data=Tuda-De + SWC (min. pruned), Vocabulary=350,029, LM=4-gram KN + 2-layer LSTM LM2018.0713.14
TransformerTraining Datasets=TD, SW, CV, Training data hours (h)=986, LM=RNNLM (4 x 1024)2020.0714.1
TransformerTraining Datasets=TD, SW*, LV*, Training data hours (h)=1460, LM=RNNLM (2 x 650), Alignment=CTC-segmentation2020.0714.3
TDNN-HMMTraining Datasets=TD, SW, CV, Training data hours (h)=649, LM=4-gram KN2020.0714.8
TDNN-HMMTraining Datasets=TD, SW, MA, Training data hours (h)=412, LM=4-gram KN2020.0715.3
TDNN-HMMTraining Data=Tuda-De + SWC (min. pruned), Vocabulary=350,029, LM=4-gram KN2018.0715.32
TransformerTraining Datasets=TD, SW, CV, Training data hours (h)=986, LM=RNNLM (2 x 650)2020.0716
TransformerTraining Datasets=TD, SW, CV, Training data hours (h)=649, LM=RNNLM (2 x 650)2020.0716.4
TDNN-HMMTraining Data=Tuda-De + SWC (min. pruned), Vocabulary=182,784, LM=3-gram KN2018.0718.39
TransformerTraining Datasets=TD, SW*, LV*, Training data hours (h)=1460, LM=None, Alignment=CTC-segmentation2020.0719.3
TDNN-HMMTraining Data=Tuda-De + SWC (min. pruned), Vocabulary=126,794, LM=3-gram KN2018.0720.3
TDNN-HMMTraining Data=Tuda-De + SWC (cons. pruned), Vocabulary=126,794, LM=3-gram KN2018.0720.91
GMM-HMMTraining Data=Tuda-De + SWC (min. pruned), Vocabulary=350,029, LM=4-gram KN2018.0724.91
GMM-HMMTraining Data=Tuda-De + SWC (min. pruned), Vocabulary=182,784, LM=3-gram KN2018.0726.92
TDNN-HMMTraining Data=Tuda-De, Vocabulary=126,794, LM=3-gram KN2018.0728.08
GMM-HMMTraining Data=Tuda-De + SWC (min. pruned), Vocabulary=126,794, LM=3-gram KN2018.0729.79
GMM-HMMTraining Data=Tuda-De + SWC (cons. pruned), Vocabulary=126,794, LM=3-gram KN2018.0729.97
TDNN-HMMTraining Data=Tuda-De, Vocabulary=28,131, LM=3-gram KN2018.0735.53
GMM-HMMTraining Data=Tuda-De, Vocabulary=126,794, LM=3-gram KN2018.0737.47
GMM-HMMTraining Data=Tuda-De, Vocabulary=28,131, LM=3-gram KN2018.0745.31