Reproduce “Attention-Based Models for Speech Recognition” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jan Chorowski, Dzmitry Bahdanau, Dmitriy Serdyuk, Kyunghyun Cho, Yoshua Bengio
The authors propose an attention-based phoneme recognizer that combines acoustic features with convolutional features from the previous alignment and smooth-focus normalization. On TIMIT, test-set phoneme error rate falls from 18.7% to 17.6%, best among the paper’s variants but above a 16.7% HMM/convolutional comparator. Long-input decoding reaches about 20% on constructed concatenations with a restricted attention window.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.