PytorchSR Save

Implementation Speech Recognition Papers

Authors

1. appleholic ([email protected])
   • Occupation KakaoBrain AI Developer, june.one

Implementations

sources

1. Completed to code training template.
2. TODOS:
   • evaluate
   • more papers

References

• phoneme classification

1. Prenet + CBHG in paper:
   • paper: Tacotron: Towards End-to-End Speech Synthesis
   • PER : to be reported
2. Revising GRU :
   • paper: Improving speech recognition by revising gated recurrent units
   • Minimal GRU:
     • Implemented No Reset Gate GRU
     • TODOs:
       • Cuda Base Implementation
         • reference impl sample link : https://github.com/chrischoy/pytorch-custom-cuda-tutorial
       • Recurrent Dropout

Setup and Run

Environment

• python 3.6
• pytorch 0.3.1
• hyperparameters with yaml (in hparams folder)

Setup

$ pip install -r requirements.txt

Command

$ python run.py train --model cbhg

Purposes:

1. Study Speech Recognition Systems
2. Well define source code structure in using pytorch