This is a project made with Elisa Cascina for the Data Science and Engineering course Data Science Lab: Process and Methods in PoliTo.
In this project we introduce a possible approach for classification problem in the field of natural language processing and speech recognition. In particular, the proposed approach is based on the Mel-frequency Cepstral Coefficients (MFCC), a widely used technique for extracting features from the audio signal. Every mfcc is divided into blocks in the time domain, some summary statistics are computed and used as input for a Random Forest classification model. The presented method overcomes the benchmark set for the issue and delivers overall acceptable outcomes.
You can find an in-depth explaination in this report.
Install requirements:
pip3 install -r requirements.txtDownload the dataset here and move the folder 'dsl_data' into the project folder.