The present repository aims at presenting a method allowing to perform emotion classification on the EmoDB dataset. This dataset contains samples of speech presenting one of 7 emotions: happiness, fear, anger, sadness, disgust, neutral or bordom.
No particular prerequisites except docker.
In order to build the docker image and download the data, run build.sh from the project root directory.
Once this is done, run run.sh to start a docker container.
The jupyter lab instance and the flask API are respectively exposed on ports 8887 and 5000.
In order to access the jupyter lab instance, you will need to copy-paste the token displayed on the output of the shell where you ran run.sh for security reason.
In order to access the GUI of the flask API, just open a tab on the 5000 port.
.
├── src # Source files
│ ├── `misc_funcs.py` # Miscellanous functions for data preparation/cleaning
│ ├── `model.py` # Classes and methods to perform the proper modelling
│ ├── `visualisation.py` # Methods to visualize the datas
│ ├── `preprocess.py` # Preprocessing methods (e.g. data augmentation etc)
├── lib # Modified libraries the project depends on
│ ├── pytorch-cnn-viz # Allows to generate class maximum output images
├── models # directory where we save the fitted models(.pkl,.json)
│ ├── weights # directory where we save the weights for CNN_classif objects
├── data
│ ├── mfcc # mfcc converted files from wav directory
│ ├── wav # wav files from the EmoDB dataset
├── docs # documentations
│ ├── Presentation.pdf # Presentation of the project/results
│ ├── Source.pdf # Main source of the project
├── results # results images
├── `Report_SER.ipynb` # Complete analysis and results of the project
├── static # CSS files for the GUI
├── `build.sh` # Bash executable to build the docker image
├── templates # HTML files for the GUI
├── `requirements.txt` # required libs for docker
└── `README.md`
A GUI shows up if you open the http://127.0.0.1:5000/: enter the filename in the main field to obtain the prediction of the most recently compute d model. TrainModel
URL: /trainmodel
Method:: GET
URL Params: n_epoch=[integer]: number of epochs to run the model for
Success Response Code: 200
Error Response Code: 201
Return: None
**Sample Call: **curl http://127.0.0.1:5000/trainmodel?nb_epoch=1
URL: /predict
Method:: POST
URL Params: filename=[string]: name of the file to receive a prediction for.
Success Response Code: 200
Error Response Code: 201
Return: { "predicted": <derp predicted_emotion>, "true_label": "<true_emotion>" }
**Sample Call: ** curl -X POST http://127.0.0.1:5000/predict?filename=03a07La
URL: /predict
Method:: POST
URL Params: None
Data Params: json file associating string values corresponding to the filenames to some keys. Example:
{
"key1":"03a04Wc",
"key2":"12a02Ec",
...
}
Success Response Code: 200
Error Response Code: 201
Return:
{ "key1": { "predicted": "<predicted_emotion1>", "true_label": "<true_emotion1>" },
"key2": { "predicted": <predicted_emotion2>, "true_label": "<true_emotion2>" },
... }
Sample Call: curl -X POST -d @data.json http://127.0.0.1:5000/predictJSON --header "Content-Type:application/json"