Skip to content

this projects contains set of codes that will process the raw audio data for dataset creation and build MLP and CCN models forreal-time voice recognition with 4 classes. at last we implemented this model on a simple game

License

Notifications You must be signed in to change notification settings

mohammadreza-babaeimosleh/real-time_audio_proceessing_on_game

Repository files navigation

Audio processing using MLP and CNN artichectures

This project contains codes to process the raw audio data for dataset creation and build MLP and CCN models for real-time voice recognition with 4 classes. At last, we implemented this model on a simple game

Dataset

we have not used any clean and pre-built dataset for this project instead of that we used our self-made dataset. you can see the codes for preprocessing and also data augmentation in related notebooks
this dataset contains 1628 audio folders(originally 814) that are separated into 4 groups: up, down, right, and left also it contains 6 .csv file (2 for each train, test, and validation) that are created using codes that are available for you to see
it is worth mentioning that labels are encoded using ono-hot-encoding

Models

in this project we used two different types of architecture for audio processing: MLP and CNN
As a result of our noisy and small dataset, the MLP model works well in theoretical research but fails to classify in real-time experiments. So we used ensemble learning as a method to increase our model's real-time accuracy and it worked quite satisfiying
CNN model just worked well and you can see the results in the codes mentioned the only drawback of it was that has a high computational cost and it will slow down our game

About

this projects contains set of codes that will process the raw audio data for dataset creation and build MLP and CCN models forreal-time voice recognition with 4 classes. at last we implemented this model on a simple game

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published