https://github.com/jingyonghou/RPN_KWS
Region proposal network based small-footprint keyword spotting (Pytorch)
https://github.com/jingyonghou/RPN_KWS
Last synced: 7 months ago
JSON representation
Region proposal network based small-footprint keyword spotting (Pytorch)
- Host: GitHub
- URL: https://github.com/jingyonghou/RPN_KWS
- Owner: jingyonghou
- License: mit
- Created: 2019-07-29T19:17:08.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2023-11-15T01:31:37.000Z (almost 2 years ago)
- Last Synced: 2025-03-31T17:55:02.456Z (7 months ago)
- Language: Python
- Homepage:
- Size: 138 KB
- Stars: 54
- Watchers: 2
- Forks: 16
- Open Issues: 4
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- awesome-keyword-spotting - Region Proposal Network Based Small-Footprint Keyword Spotting
README
# RPN_KWS
Pytorch code of paper "Region Proposal Network Based Small-Footprint Keyword Spotting"
https://ieeexplore.ieee.org/document/8807313
Please cite the work below if you want to use the code or want to do research related to our work
```
@ARTICLE{hou2019region,
author={Hou, Jingyong and Shi, Yangyang and Ostendorf, Mari and Hwang, Mei-Yuh
and Xie, Lei },
journal={IEEE Signal Processing Letters},
title={Region Proposal Network Based Small-Footprint Keyword Spotting},
year={2019},
volume={26},
number={10},
pages={1471-1475}
}
```
I will release a new version of RPN KWS with an Online Hard Example Mining (OHEM) algorithm, which will improve our system.
https://github.com/jingyonghou/RPN_KWS_OHEM
## Detection samples
Selected two utterances which contains predefined keyword. The red box is the ground-truth start-end area of keyword from forced-alignment, the blue box is the best anchor selected according to the classification score, the green box is the proposed region proposal corresponding to the best anchor.
## Running environment
### Python 2.7.15
### pytorch 0.4.1
### CUDA 8.0 or higher
### Kaldi
You should know basic knowledge of Kaldi before looking at the run script. I use Kaldi to extract Fbank features and do a global CMVN using the statictics from all training set. You should add cmd.sh, path.sh, steps and utils to your working dir before you run the script.
### Please follow the run_rpn_kws.sh script to learn how to run the code
## reference
https://github.com/jwyang/faster-rcnn.pytorch
https://github.com/vesis84/kaldi-io-for-python