Digital Sound module2

Introduction

This project is a module #2 of Digital Sound lecture: Speaker Recognition

Learning Objective

Implementing voice recognition that is used in AI speakers

Content

Topic

Recognizing speaker’s voice

Principle & Theory

Distinguishing Speaker’s voice patterns in that every person have different voice feature

Problem

We want to provide personal service for each family member by AI speaker using Speaker Recognition

Setting data

• Collect 12 students’ audio data. Each of the audio data has to read the given script and record with coolEdit(WAV, 16000Hz, 16-bit, mono).
• After collecting the data, use them as training data
• All data will be shared with classmates

All data that have used for training models will NOT be OPENED to PUBLIC due to privacy issues

Solution

I have looked up the code from this blog and applied to my project
Image down below is a flow chart how the code works in total

Code Description

train_model.py

• Input: folder name of training data (folder has to exists in executing directory)
• Output: speaker_name.gmm

test_model.py

• Input: total number of test data and folder name of test data (folder has to exists in executing directory)
• Output: identified speaker names

show_all_graphs.py
Mel Spectogram, MFCC, GMM graphs.

Minor issues

Library

Mainly used sklearn and speakerfeatures with Python