Reproduce “Speech Recognition with Deep Recurrent Neural Networks” 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

Alex Graves, Abdel-rahman Mohamed, Geoffrey Hinton

The authors train recurrent networks to recognize phonemes from speech without supplied frame-level alignments, using connectionist temporal classification and stacked bidirectional long short-term memory layers to combine depth with past and future context. On TIMIT’s 24-speaker core test set, a pretrained three-layer transducer achieves 17.7% phoneme error rate. This is phoneme recognition, not large-vocabulary or streaming recognition; experiments ran once.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.