All the code and the datasets are available at the following repository: https://github.com/gyawalisaurav/AudioLocalization
The repository contains three matlab files:
- GetRIR.m
- DataCollection.m
- DataProcessing.m
GetRIR is a helper function to collect data. It can simply be called in Matlab without any arguments to generate the room impulse response. Additionally it returns the time vector for the impulse response and also the Discrete Fourier Transform of the raw audio input. Here is an example call to this funtion:
[time_vector, RIR, fourier_transform] = GetRIR();
DataCollection file asks for the room name and gathers the specified number of RIR samples. This sample is stored as matlab matrix and later saved as a file that can be collected by matlab, python or many other tools. It can simply be called without any arguments.
DataProcessing file searches for any mat files in the directory and processes the impulse responses stored in the mat files. It extracts important features from the data as well and creates a combined feature and labels matrix that can be fed into any machine learning framework.
After the variable with features is set in the workspace, the classification learner app in matlab can be used to train a SVM. The room values/labels should be used as response in the training data set. With n-fold cross validation, SVM can be trained and the resulting classifier can be exported to be used for classifying rooms in real life situations. The RIR can simply be passed to the exported classifier to obtain the predicted label.