Automatic Speech Recognition on Hub5'01 (test)
5.2WER (SWB)Single headed attention based sequence-to-sequence model
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Single headed attention based sequence-to-sequence modelTraining Data=2k-lrg, External Language Model=true, Cross-utterance=true2020.01 | 5.2 | — | 6.3 | 9.7 | |
| Single headed attention based sequence-to-sequence modelTraining Data=2k-lrg, External Language Model=false, Cross-utterance=false2020.01 | 5.5 | — | 6.7 | 10.3 | |
| Single headed attention based sequence-to-sequence modelTraining Data=2k, External Language Model=true, Cross-utterance=true2020.01 | 6.5 | — | 8.2 | 11.4 | |
| Single headed attention based sequence-to-sequence modelTraining Data=2k, External Language Model=true, Cross-utterance=false2020.01 | 6.6 | — | 8.3 | 11.5 | |
| Single headed attention based sequence-to-sequence modelTraining Data=300h, External Language Model=true, Cross-utterance=true2020.01 | 6.8 | — | 9.1 | 13.4 | |
| Single headed attention based sequence-to-sequence modelTraining Data=2k, External Language Model=false, Cross-utterance=false2020.01 | 6.8 | — | 8.7 | 11.7 | |
| Single headed attention based sequence-to-sequence modelTraining Data=300h, External Language Model=true, Cross-utterance=false2020.01 | 7 | — | 9.3 | 13.8 | |
| Single headed attention based sequence-to-sequence modelTraining Data=300h, External Language Model=false, Cross-utterance=false2020.01 | 8.1 | — | 11 | 15.7 | |
| attention modelLM=none, label unit=BPE 10K2018.05 | — | 19.9 | — | — | |
| attention modelLM=none, label unit=BPE 1K2018.05 | — | 19.7 | — | — | |
| attention modelLM=LSTM, label unit=BPE 1K2018.05 | — | 18.1 | — | — | |
| hybridLM=4-gram, label unit=CDp2018.05 | — | 14.7 | — | — | |
| hybridLM=LSTM, label unit=CDp2018.05 | — | 12.9 | — | — |