End-to-end ASR: from Supervised to Semi-Supervised Learning with Modern Architectures

Open paper
Benchmark Model Rank Results
speech-recognition-on-librispeech-test-cleanConv + Transformer AM + Pseudo-Labeling (ConvLM with Transformer Rescoring)#74Word Error Rate (WER): 2.03
speech-recognition-on-librispeech-test-cleanConv + Transformer AM (ConvLM with Transformer Rescoring) (LS only)#89Word Error Rate (WER): 2.31
speech-recognition-on-librispeech-test-otherConv + Transformer AM (ConvLM with Transformer Rescoring)#62Word Error Rate (WER): 4.11
speech-recognition-on-librispeech-test-otherConv + Transformer AM (ConvLM with Transformer Rescoring) (LS only)#86Word Error Rate (WER): 5.18

Rank counts only results with a code link.