Speech Recognition On Switchboard Hub500

评估指标

Percentage error

评测结果

各个模型在此基准测试上的表现结果

Paper TitleRepository
Deep Speech20Deep Speech: Scaling up end-to-end speech recognition
DNN-HMM18.5--
CD-DNN16.1--
DNN16Building DNN Acoustic Models for Large Vocabulary Speech Recognition
DNN + Dropout15Building DNN Acoustic Models for Large Vocabulary Speech Recognition
DNN MMI12.9--
HMM-TDNN + pNorm + speed up/down speech12.9--
DNN MPE12.9--
DNN BMMI12.9--
Deep Speech + FSH12.6Deep Speech: Scaling up end-to-end speech recognition
HMM-DNN +sMBR12.6--
CNN + Bi-RNN + CTC (speech to letters), 25.9% WER if trainedonlyon SWB12.6Deep Speech: Scaling up end-to-end speech recognition
DNN sMBR12.6--
Deep CNN (10 conv, 4 FC layers), multi-scale feature maps12.2Very Deep Multilingual Convolutional Neural Networks for LVCSR-
CNN11.5--
HMM-TDNN + iVectors11--
CNN on MFSC/fbanks + 1 non-conv layer for FMLLR/I-Vectors concatenated in a DNN10.4--
HMM-TDNN trained with MMI + data augmentation (speed) + iVectors + 3 regularizations + Fisher (10% / 15.1% respectively trained on SWBD only)9.2--
HMM-BLSTM trained with MMI + data augmentation (speed) + iVectors + 3 regularizations + Fisher8.5--
IBM 20158.0The IBM 2015 English Conversational Telephone Speech Recognition System-
0 of 30 row(s) selected.
Speech Recognition On Switchboard Hub500 | SOTA | HyperAI超神经