https://github.com/desh2608/kaldi-noise-vectors
Implementation of different noise embeddings for noise aware training of Kaldi acoustic models.
https://github.com/desh2608/kaldi-noise-vectors
Last synced: about 2 months ago
JSON representation
Implementation of different noise embeddings for noise aware training of Kaldi acoustic models.
- Host: GitHub
- URL: https://github.com/desh2608/kaldi-noise-vectors
- Owner: desh2608
- License: mit
- Created: 2021-02-02T15:27:02.000Z (over 4 years ago)
- Default Branch: main
- Last Pushed: 2021-02-13T19:13:03.000Z (over 4 years ago)
- Last Synced: 2025-03-25T07:11:24.086Z (2 months ago)
- Language: Shell
- Size: 43 KB
- Stars: 13
- Watchers: 3
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# kaldi-noise-vectors
This repository contains the baselines and proposed noise vectors from the paper:
[Frustratingly Easy Noise-aware Training of Acoustic Models](https://arxiv.org/abs/2011.02090).
### Setup
**Prerequisite:** The setup below assumes that you have a working installation of
the [Kaldi](https://github.com/kaldi-asr/kaldi) ASR toolkit.
* Copy the contents of the `src` directory to the corresponding directory in your
Kaldi installation.
* Navigate to `/path/to/kaldi/src` and run the following:
```shell
cd ivectorbin && make compute-noise-vector compute-noise-vector-seltzer && cd ..
```
* If you additionally want to use the proposed online MLE and MAP noise vectors,
run the following:
```shell
cd ivector && make online-noise-vector && cd ..
cd ivectorbin && make compute-noise-prior compute-noise-vector-online && cd ..
```
### Usage
We provide example usage on the Aurora4 dataset. For model details and how to run
the recipes, please check `egs/aurora4/s5/README.md`.
### Citation
If you found this code useful, consider citing:
```shell
@article{Raj2020FrustratinglyEN,
title={Frustratingly Easy Noise-aware Training of Acoustic Models},
author={Desh Raj and Jes{\'u}s Villalba and Daniel Povey and Sanjeev Khudanpur},
journal={ArXiv},
year={2020},
volume={abs/2011.02090}
}
```