Scribosermo: Fast Speech-to-Text models for German and other Languages

Benchmark Model Rank Results
speech-recognition-on-common-voice-englishConformerCTC-L (5-gram)#6Test WER: 9.06%
speech-recognition-on-common-voice-englishConformerCTC-L (5-gram, charbased)#7Test WER: 14.38%
speech-recognition-on-common-voice-frenchConformerCTC-L (5-gram)#2Test WER: 8.13%
speech-recognition-on-common-voice-frenchConformerCTC-L (no-LM)#6Test WER: 10.19 %
speech-recognition-on-common-voice-frenchQuartzNet15x5FR (D7)#7Test WER: 11.0%
speech-recognition-on-common-voice-frenchQuartzNet15x5FR (CV-only)#8Test WER: 12.1%
speech-recognition-on-common-voice-germanConformerCTC-L (5-gram)#3Test WER: 4.05%Test CER: 1.37%
speech-recognition-on-common-voice-germanQuartzNet15x5DE (D37, 5-gram)#8Test WER: 6.6%Test CER: 2.7%
speech-recognition-on-common-voice-germanConformerCTC-L (no LM)#10Test WER: 7.33%Test CER: 2.05%
speech-recognition-on-common-voice-germanQuartzNet15x5DE (CV-only, 5-gram)#11Test WER: 7.7%Test CER: 3.2%
speech-recognition-on-common-voice-spanishConformerCTC-L (5-gram)#4Test WER: 5.68%
speech-recognition-on-common-voice-spanishConformerCTC-L (no-LM)#6Test WER: 7.46 %
speech-recognition-on-common-voice-spanishQuartzNet15x5ES (D8)#8Test WER: 10.0%
speech-recognition-on-common-voice-spanishQuartzNet15x5ES (CV-only)#9Test WER: 10.5%
speech-recognition-on-tudaQuartzNet15x5DE (D37)#2Test WER: 10.2%