federatedSpeechCommands

KO | EN

gitlite — search

#javascript #react #go #android #machine-learning #deep-learning #linux #awesome #cli #awesome-list #security #ios #openai #open-source

federatedSpeechCommands

(★ 11)

Speech recognition with federated learning

federatedSpeechCommands 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

Convolutional neural networks for [Google speech commands data set](https://research.googleblog.com/2017/08/launching-speech-commands-dataset.html)
with [PyTorch](http://pytorch.org/).

# With federated Learning

Run `federated_train_speech_commands_gpu.py` for simulating federated training with multiple clients, for example
```
python federated_train_speech_commands.py --model=vgg11 --optim=sgd --lr-scheduler=plateau --learning-rate=0.01 --lr-scheduler-patience=5 --max-epochs=1 --batch-size=156 --clients=2 --matrix-size=500
```
```
CUDA_VISIBLE_DEVICES=0,1 python federated_train_speech_commands_cpu_v5_collect_gradient.py --model=conv --optim=sgd --lr-scheduler=plateau --learning-rate=0.01 --lr-scheduler-patience=5 --max-epochs=1 --batch-size=128 --clients=3 --matrix-size=100  --num-threads=10
```

# General
We, [xuyuan](https://github.com/xuyuan) and [tugstugi](https://github.com/tugstugi), have participated
in the Kaggle competition [TensorFlow Speech Recognition Challenge](https://www.kaggle.com/c/tensorflow-speech-recognition-challenge)
and reached the 10-th place. This repository contains a simplified and cleaned up version of our team's code.

# Features
* `1x32x32` mel-spectrogram as network input
* single network implementation both for CIFAR10 and Google speech commands data sets
* faster audio data augmentation on STFT
* Kaggle private LB scores evaluated on 150.000+ audio files

# Results
Due to time limit of the competition, we have trained most of the nets with `sgd` using `ReduceLROnPlateau` for 70 epochs.
For the training parameters and dependencies, see [TRAINING.md](TRAINING.md). Earlier stopping the train process will sometimes produce a better score in Kaggle.

<table><tbody>
<th valign="bottom">&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;Model&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;</th>
<th valign="bottom">CIFAR10 test set accuracy</th>
<th valign="bottom">Speech Commands test set accuracy</th>
<th valign="bottom">Speech Commands test set accuracy with crop</th>
<th valign="bottom">Speech Commands Kaggle private LB score</th>
<th valign="bottom">Speech Commands Kaggle private LB score with crop</th>
<th valign="bottom">&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;Remarks&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;</th>

<tr>
<td align="left">ResNet32</td>
<td align="center">-</td>
<td align="center">96.181419%</td>
<td align="center">96.196050%</td>
<td align="center">0.87078</td>
<td align="center">0.87419</td>
<td align="left"></td>
</tr>

<tr>
<td align="left">WRN-28-10-dropout</td>
<td align="center">96.22%</td>
<td align="center">97.702999%</td>
<td align="center">97.717630%</td>
<td align="center">0.89580</td>
<td align="center">0.89568</td>
<td align="left"></td>
</tr>

<tr>
<td align="left">WRN-52-10</td>
<td align="center">-</td>
<td align="center">98.039503%</td>
<td align="center">97.980980%</td>
<td align="center">0.88159</td>
<td align="center">0.88323</td>
<td align="left">another trained model has 97.52%/0.89322</td>
</tr>

<tr>
<td align="left">ResNext29 8x64</td>
<td align="center">-</td>
<td align="center">97.190929%</td>
<td align="center">97.161668%</td>
<td align="center">0.89533</td>
<td align="center">0.89733</td>
<td align="left">our best model during competition</td>
</tr>

<tr>
<td align="left">DenseNet-BC (L=100, k=12)</td>
<td align="center">95.52%</td>
<td align="center">97.161668%</td>
<td align="center">97.147037%</td>
<td align="center">0.88946</td>
<td align="center">0.89134</td>
<td align="left"></td>
</tr>

<tr>
<td align="left">DenseNet-BC (L=190, k=40)</td>
<td align="center">-</td>
<td align="center">97.117776%</td>
<td align="center">97.147037%</td>
<td align="center">0.89369</td>
<td align="center">0.89521</td>
<td align="left"></td>
</tr>

</tbody></table>

# Results with Mixup

After the competition, some of the networks were retrained using [mixup: Beyond Empirical Risk Minimization](https://arxiv.org/abs/1710.09412) by Hongyi Zhang, Moustapha Cisse, Yann N. Dauphin and David Lopez-Paz.

<table><tbody>
<th valign="bottom">&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;Model&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;</th>
<th valign="bottom">CIFAR10 test set accuracy</th>
<th valign="bottom">Speech Commands test set accuracy</th>
<th valign="bottom">Speech Commands test set accuracy with crop</th>
<th valign="bottom">Speech Commands Kaggle private LB score</th>
<th valign="bottom">Speech Commands Kaggle private LB score with crop</th>
<th valign="bottom">&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;Remarks&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;</th>

<tr>
<td align="left">WRN-52-10</td>
<td align="center">-</td>
<td align="center">97.454279%</td>
<td align="center">97.498171%</td>
<td align="center">0.90273</td>
<td align="center">0.90355</td>
<td align="left">same score as the 16-th place in Kaggle</td>
</tr>

</tbody></table>

federatedSpeechCommands

federatedSpeechCommands 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

federatedSpeechCommands

federatedSpeechCommands 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README