Automated Child Voice Generation: Methodology and Implementation

Authors: Shaimaa Alwaisi, Mohammed Salah Al-Radhi, Géza Németh

Child_Speech_Synthesis

Boy Speaker:

Text: "I THINK WHAT IS BEING SHOWN IS THE SYSTEM FOR THE HEART"

Ground_truth	Proposed system	ParallelWavGan	WorldVoc

Text: "THESE CIRCUITS ARE CLOSED AND THE LIGHT BULB SHOULD BE LIGHTING UP BUT FOR SOME REASON IT'S NOT "

Ground_truth	Proposed system	ParallelWavGan	WorldVoc

Girl Speaker:

Text: "I THINK IT WOULD GET BRIGHTER TO BREAK THE BULB OR IT WOULDN'T WORK"

Ground_truth	Proposed system	ParallelWavGan	WorldVoc

Text: "I MEAN LIKE IT POWERS MOSTLY EVERYTHING"

Ground_truth	Proposed system	ParallelWavGan	WorldVoc

Visualization

The image below demonstrates The comparison of Mel spectrograms and F0 values between the ground truth and three models: (b) AutoVocoder, (c) ParallelWavGan, and (d) WORLD Vocoder for both Boy and Girl.