Criterion suitable for the model

We supports 20+ models and is continuously updated. So we try to explain the possible criterion for each model.

Model - Criterion

1. DeepSpeech2 - ctc
2. Jasper - ctc
3. QuartzNet - ctc
4. Transformer with CTC - ctc
5. Conformer with CTC - ctc
6. Listen Attend Spell - cross_entropy, label_smoothed_cross_entropy
7. Location-aware attention based Listen Attend Spell - cross_entropy, label_smoothed_cross_entropy
8. Multi-head attention based Listen Attend Spell - cross_entropy, label_smoothed_cross_entropy
9. Speech-Transformer - cross_entropy, label_smoothed_cross_entropy
10. VGG-Transformer - cross_entropy, label_smoothed_cross_entropy
11. Conformer with LSTM Decoder - cross_entropy, label_smoothed_cross_entropy
12. Joint CTC-Attention based Transformer - joint_ctc_cross_entropy
13. Joint CTC-Attention based Listen Attend Spell - joint_ctc_cross_entropy
14. Deep CNN Encoder with Joint CTC-Attention Listen Attend Spell - joint_ctc_cross_entropy
15. RNN-Transducer - transducer
16. Transformer Transducer - transducer
17. Conformer - transducer
18. ContextNet - transducer
19. LSTM Language Model - perplexity
20. Transformer Language Model - perplexity

Criterion

1. ctc
   conformer, contextnet, deepspeech2, jasper5x3, jasper10x5, quartznet5x5, quartznet10x5, quartznet15x5, transformer_with_ctc
2. cross_entropy
   conformer_lstm, contextnet_lstm, listen_attend_spell, listen_attend_spell_with_location_aware, listen_attend_spell_with_multi_head, transformer, vgg_transformer
3. label_smoothed_cross_entropy
   conformer_lstm, contextnet_lstm, listen_attend_spell, listen_attend_spell_with_location_aware, listen_attend_spell_with_multi_head, transformer, vgg_transformer
4. joint_ctc_cross_entropy
   deep_cnn_with_joint_ctc_listen_attend_spell, joint_ctc_conformer_lstm, joint_ctc_listen_attend_spell, joint_ctc_transformer
5. transducer
   conformer_transducer, contextnet_transducer, rnn_transducer, transformer_transducer
6. perplexity
   lstm_lm, transformer_lm