Recurrent Poisson Process Unit for Speech Recognition
Authors: Hengguan Huang, Hao Wang, Brian Mak6538-6545
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental evaluations on ASR tasks including Chi ME-2, WSJ0 and WSJ0&1 demonstrate the effectiveness and benefits of the RPPU. |
| Researcher Affiliation | Academia | Hengguan Huang,1 Hao Wang,2 Brian Mak1 1The Hong Kong University of Science and Technology 2Massachusetts Institute of Technology {hhuangaj,mak}@cse.ust.hk, hwang87@mit.edu |
| Pseudocode | No | The paper describes mathematical formulations and update rules for its model, but it does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not include any statements about releasing code, nor does it provide links to a code repository for the described methodology. |
| Open Datasets | Yes | Datasets We evaluated the proposed RPPU on three ASR corpora: Chi ME-2 (Vincent et al. 2013), WSJ0 (Garofolo et al. 1993) and WSJ0&1 (Garofolo et al. 1993; Consortium and others 1994). |
| Dataset Splits | Yes | The training set contains about 15 hours of speech with 7138 simulated noisy utterances. The transcriptions are based on those of the WSJ0 training set. The development and test sets contain 2460 and 1980 simulated noisy utterances, respectively. |
| Hardware Specification | Yes | The timing experiments used the Theano package and were performed on a machine running the Ubuntu operating system with a single Intel Core i7-7700 CPU and a GTX 1080Ti GPU. |
| Software Dependencies | No | The paper mentions the use of 'Theano package' and 'Kaldi s5 recipe' but does not specify version numbers for these or other software dependencies. |
| Experiment Setup | Yes | We applied a dropout rate of 0.1 to the connections between recurrent layers. The learning rate for LSTM/SRU, Quasi-RNN and RPPU models was initially set to 0.25, 0.2, and 0.07 respectively. We decayed the learning rate until it went below 1 10 6. ... For simplicity, in the intensity function of RPPU, c was set to 100 and ϵ was set to 0.01. |