This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. In addition a 6gram KenLM model was trained and used. The KenLM model was based on train+validation Common Voice 8 It achieves results depicted on the rigth side on the model card (testset CV8)
Dutch wav2vec2-xls-r-300m model using Common Voice 8 dataset
More information needed
The model was trained on Dutch common voice 8 with 75 epochs. The train set consisted of the common voice 8 train set and evaluation set was the common voice 8 validation set. The WER reported is on the common voice 8 test set which was not part of training nor validation (eval)
pytorch_model.bin 1.3 GBlanguage_model/6gram.bin 36 MBtraining_args.bin 3.1 kB