Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/blueloveth/speech_commands_recognition
CCF练习赛-通用音频分类
https://github.com/blueloveth/speech_commands_recognition
Last synced: 3 months ago
JSON representation
CCF练习赛-通用音频分类
- Host: GitHub
- URL: https://github.com/blueloveth/speech_commands_recognition
- Owner: blueloveTH
- Created: 2020-11-19T15:49:30.000Z (about 4 years ago)
- Default Branch: master
- Last Pushed: 2021-05-31T03:16:18.000Z (over 3 years ago)
- Last Synced: 2023-04-16T03:20:59.027Z (over 1 year ago)
- Language: Jupyter Notebook
- Homepage:
- Size: 511 KB
- Stars: 13
- Watchers: 2
- Forks: 5
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Speech Commands Recognition
## 内容介绍
https://zhuanlan.zhihu.com/p/331833198
## 实验结果
| Local CV Score | Test Score |
| -------------- | ------------- |
| 0.977 ± 0.001 | 0.975 ± 0.001 |
本方案基于pytorch和[keras4torch](https://github.com/blueloveTH/keras4torch)。为方便移植到其他框架测试，下面列出了训练用到的主要设定。
## 主要设定
| setting | value |
| ----------------- | ---------------------------- |
| features | 1x32x32 melspectrogram |
| model | wide resnet28 |
| total parameters | 36491726 |
| epochs | 40 |
| batch size | 96 |
| optimizer | SGD with momentum |
| learning rate | 1e-2 -> 3e-3 -> 9e-4 -> 8e-5 |
| L2 regularization | 1e-2 |
| label smoothing | 0.1 |
| epoch time | 82s (1 * RTX 2080Ti) |
## 模型结构
## 运行仓库代码
#### 环境配置
```txt
torch>=1.6.0
keras4torch==1.1.3
scikit-learn==0.23.2
librosa==0.8.0
```
如果使用linux系统，需要先执行如下命令才能安装librosa。
```bash
! sudo apt-get install -y libsndfile1
```
#### 数据预处理
确保原始数据被放在data/ 文件夹中，运行preprocess.ipynb。
这些文件的结构如下：
- data/
- train/
- test/
- preprocess.ipynb
- train.ipynb
#### 训练和预测
在上一步完成的基础上，运行train.ipynb。
结束后，对测试集的预测（概率值）将被保存为一个.npy文件。
## 问题反馈
+ [Github Issue](https://github.com/blueloveTH/speech_commands_recognition/issues)
+ Email: [email protected]