Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/drukka/command-words-recognition-keras
Speech recognition of command words using Keras and Tensorflow2
https://github.com/drukka/command-words-recognition-keras
Last synced: 2 months ago
JSON representation
Speech recognition of command words using Keras and Tensorflow2
- Host: GitHub
- URL: https://github.com/drukka/command-words-recognition-keras
- Owner: drukka
- License: mit
- Archived: true
- Created: 2019-04-23T10:55:00.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2019-04-24T09:42:47.000Z (over 5 years ago)
- Last Synced: 2024-08-02T19:01:53.889Z (6 months ago)
- Language: Python
- Homepage:
- Size: 12.7 KB
- Stars: 2
- Watchers: 1
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE.md
Awesome Lists containing this project
- Awesome-Tensorflow2 - drukka/command-words-recognition-keras
README
CNN - Command Words Recognition
=========
Speech recognition in Python using [Tensorflow 2](https://www.tensorflow.org/alpha) and [Keras](https://www.tensorflow.org/versions/r2.0/api_docs/python/tf/keras) high level API.
Convolutional Neural Networks (__CNN__) were invented to classify time invariant data (like images). Recent researches found that, if sound is converted into its [spectrogram](https://en.wikipedia.org/wiki/Spectrogram) (or better: [Log-Mel spectrogram](https://en.wikipedia.org/wiki/Mel-frequency_cepstrum), then convolutional neural networks can be applied on sound's features (aka spectrogram) for training speech recognition model.
## Usage
1. Clone repository:
```cli
cd /path/to/these/files/
git clone https://github.com/tyukesz/command-words-recognition-keras
```
2. Check __constants.py__ file for parameters like:
* DATASET_URL // URL where ZIPed sounds files are stored
* SAMPLE_RATE // Recomended 16KHz
* EPOCHS // number of training epochs
* BATCH_SIZE //use power of 2: 64, 128, 256, 512, ...
* TESTING and VALIDATION percentage // Recomended 15% - 15%
* WANTED_WORDS // list of command words: ['yes', 'no', 'up', 'down', 'on', 'off', ...]
* VERBOSITY //0=disable, 1=progressbar
3. Install requirements:
```cli
pip install requirements.txt
```
4. Run training on CPU:
```cli
python train.py
```
* If your are running training for first time it's recommended using '_--force_extract=True_' argument:
```cli
python train.py --force_extract=True
```
* If your __SOUNDS_DIR__ is empty it will download and extract the sounds files from provided __DATASET_URL__.
* Forcing MFCC feature extraction (_force_extract=True_) causes saving sounds features in __MFCCS_DIR__ as tensors.
* If features are already extracted, then the training begins faster since no need to recompute MFCCs of sounds, just load them as tensor.
5. \(Optional) You can load the pretrained model for [_transfer learning_](https://en.wikipedia.org/wiki/Transfer_learning):
```cli
python train.py --load_model=name_of_your_model
```
6. \(Optional) Test your model prediction with WAV file:
```cli
python predict_wav.py --load_model=name_of_your_model --wav_path=/path/to/yes.wav --num_of_predictions=2
```
* The above command will output something like:
* yes (97%)
* left (0.84%)
## CNN architecure
Using sequential Keras model with following layers:
Layer (type) | Output Shape | Params
:--- | :--- | :---
conv2d (Conv2D) | (512, 64, 32, 16) | 416
max_pooling2d (MaxPooling2D) | (512, 32, 16, 16) | 0
conv2d_1 (Conv2D) | (512, 32, 16, 32) | 4640
max_pooling2d_1 (MaxPooling2)| (512, 16, 8, 32) | 0
dropout (Dropout) | (512, 16, 8, 32) | 0
| batch_normalization_v2 (BatchNormalization) | (512, 16, 8, 32) | 128
conv2d_2 (Conv2D) | (512, 8, 4, 64) | 18496
conv2d_3 (Conv2D) | (512, 8, 4, 128) | 73856
max_pooling2d_2 (MaxPooling2)| (512, 4, 2, 128) | 0
dropout_1 (Dropout) | (512, 4, 2, 128) | 0
flatten (Flatten) | (512, 1024) | 0
dropout_2 (Dropout) | (512, 1024) | 0
dense (Dense) | (512, 256) | 262400
dense_1 (Dense) | (512, 6) | 1542
| ____________________________|
| Total params: 361,478|
| Trainable params: 361,414|
| Non-trainable params: 64|
## License
This project is licensed under the MIT License - see the [LICENSE.md](LICENSE.md) file for details.
[](https://opensource.org/licenses/MIT)