cmsfw-github

Attention-Based Models for Speech Recognition has a sharpening feature which selects a few of hidden states to attend. However, there are little details about how to select in the paper...

在dataset/icdar2013_dataset.py第93行附近 if confidence