AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Recurrent Poisson Process Unit for Speech Recognition

Hengguan Huang, Hao Wang, Brian Mak

AAAI 2019 · Architectures and efficiency › RNNs and normalization · approach: Efficient recurrent architectures

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Replace SRU/LSTM/quasi-RNN with RPPUhybrid RNN-HMM acoustic model · CHiME-2 testWER (%): 26.1 -> 24.4Improved
Replace LSTM with RPPUhybrid RNN-HMM · WSJ0 eval92-5kWER (%): 2.8 -> 2.5Improved
Replace LSTM with RPPUhybrid RNN-HMM · WSJ0&1 dev93-20kWER (%): 7.4 -> 6.9Improved
Replace LSTM with RPPUhybrid RNN-HMM · WSJ0&1 eval93-20kWER (%): 6.8 -> 6.2Improved
Replace LSTM with RPPU at low SNRhybrid RNN-HMM · CHiME-2 test, -6 dBWER (%): 42.4 -> 39.9Improved
Replace LSTM with RPPU at high SNRhybrid RNN-HMM · CHiME-2 test, 9 dBWER (%): 15.3 -> 13.2Improved
Use RPP-estimated time points to re-align frame-level targetsphoneme-level alignment · CHiME-2 devalignment similarity (%): 55.5 -> 64.7Improved
Add regularization term on intensity (weight gamma around 0.08) (ablation)RPPU · CHiME-2 devWER (%)Improved
Increase LSTM depth beyond three layers (ablation)LSTM · CHiME-2WER (%)No clear effect
Replace LSTM with SRU (12 layers)hybrid RNN-HMM · CHiME-2 testWER (%): 26.1 -> 26.2No clear effect
Replace LSTM with SRU for training speedhybrid RNN-HMM · CHiME-2training time per epoch (hr): 0.71 -> 0.32Improved
Replace SRU with RPPUSRU · CHiME-2training time per epoch (hr): 0.32 -> 0.37Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy