patrickvonplaten/wav2vec2-base-100h-2nd-try

The model was trained on librispeech-clean-train.100 with following hyper-parameters . The model is trained on two GPUs Titan RTX with total update steps 11,000 steps . Results are similar to the ones reported in https://huggingface.co/facebook/Wav2vec2-base-100h.100h . Model was trained with linear decaying learning rate with 3000 warmup steps . The result (WER) on WAV2vec-V2vec 2-base.2 (-1.6%) and (WAV2-v2vec) 6.2 - 11.2% - 1.6% - 2% - 3,