Kirili4ik/QuartzNet-ASR-pytorch

Automatic Speech Recognition (ASR) model QuartzNet trained on English CommonVoice. In PyTroch with CTC loss and beam search.

QuartzNet (ASR, 1D separable convolutions)

Model described in Kriman et al., 2019 (QuartzNet: Deep Automatic Speech Recognition with 1D Time-Channel Separable Convolutions).

Data can be downloaded here: https://commonvoice.mozilla.org/en/datasets

More files for running here (sorted indexes, preprecessed tsv, model weights): https://yadi.sk/d/tT-N6DRHkB5XTw?w=1