CNN + Bi-RNN + CTC (speech to letters), 25.9% WER if trainedonlyon SWB

Reported on 2 benchmarks across 1 task · 1 paper · 1 SOTA

Note: results are matched by exact model name. Different papers may use the same name for different model variants.

Audio2 results

Speech Recognitiononswb_hub_500 WER fullSWBCH
Percentage error· 2014-12-17
16
best: 6.8 (IBM (LSTM+Conformer encoder-decoder))
SOTA
Deep Speech: Scaling up end-to-end speech recognition arXiv:1412.5567
Speech RecognitiononSwitchboard + Hub500
Percentage error· 2014-12-17
12.6
best: 4.3 (IBM (LSTM+Conformer encoder-decoder))
Deep Speech: Scaling up end-to-end speech recognition arXiv:1412.5567