DataScienceProject

Context:

Class: Data Science Lab: process and methods, Politecnico di Torino (2022/2023)

1st year of Master's Degree in Data Science and Engineering coursework

Abstract:

This report presents a possible solution to a classification problem regarding intent detection from audio recordings. This solution is based on the spectral analysis of audios, performed by means of a spectrogram that is split into a defined number of blocks. Each block is then used to extract a summary of statistical properties used by the classification model. The result will be evaluated in terms of accuracy of the predicted intent of each recording.

Files:

• FinalProject.ipynb is a notebook in which the whole pipeline (data exploration & preprocessing, model selection, model training, prediction) is contained;
• SpokenLanguageIntentDetection.pdf is the complete paper explaining goals, reasoning and results of the project.