CR-CTC: Consistency regularization on CTC for improved speech recognition

Benchmark Model Rank Results
speech-recognition-on-aishell-1Zipformer+CR-CTC (no external language model)#5Word Error Rate (WER): 4.02Params(M): 66.2
speech-recognition-on-gigaspeech-testZipformer+pruned transducer w/ CR-CTC (no external language model)#44Word Error Rate (WER): 10.03
speech-recognition-on-gigaspeech-testZipformer+CR-CTC/AED (no external language model)#45Word Error Rate (WER): 10.07
speech-recognition-on-gigaspeech-testZipformer+pruned transducer (no external language model)#46Word Error Rate (WER): 10.2
speech-recognition-on-gigaspeech-testZipformer+CR-CTC (no external language model)#47Word Error Rate (WER): 10.28
speech-recognition-on-librispeech-test-cleanZipformer+pruned transducer w/ CR-CTC (no external language model)#60Word Error Rate (WER): 1.88
speech-recognition-on-librispeech-test-cleanZipformer+CR-CTC (no external language model)#73Word Error Rate (WER): 2.02
speech-recognition-on-librispeech-test-otherZipformer+pruned transducer w/ CR-CTC (no external language model)#56Word Error Rate (WER): 3.95
speech-recognition-on-librispeech-test-otherZipformer+CR-CTC (no external language model)#68Word Error Rate (WER): 4.35