# Low word error rate (WER)

Vietnamese automatic speech recognition model based on wav2vec 2.0 architecture, trained on 13,000 hours of unlabeled audio and 250 hours of labeled data

A 300M-parameter speech recognition model fine-tuned on the VoxRex feature extractor, optimized for Nynorsk (New Norwegian), achieving a WER of 12.22% on the NPSC test set

A speech recognition model fine-tuned with Portuguese data from public speech datasets, based on Facebook's wav2vec2-large-xlsr-53 model