Improved Noisy Student Training for Automatic Speech Recognition

Benchmark Model Rank Results
speech-recognition-on-librispeech-test-cleanContextNet + SpecAugment-based Noisy Student Training with Libri-Light#51Word Error Rate (WER): 1.7
speech-recognition-on-librispeech-test-otherContextNet + SpecAugment-based Noisy Student Training with Libri-Light#36Word Error Rate (WER): 3.4