Child Speech Synthesis using Deep Learning

Siddique, Zeba Qamaruddin (2022) Child Speech Synthesis using Deep Learning. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (2MB) | Preview
Preview	PDF (Configuration manual) Download (4MB) | Preview

Abstract

Speech synthesis or text-to-speech automatically translates the input text into spoken speech. TTS technologies have progressively increased to generate synthesized speech that is intelligible, natural, and human-sounding. Whilst extensive research on TTS has been conducted using the adult speech corpus, little or minimum investigation has been done on the synthesis of child speech. This study proposes a TTS model comprising of a pre-trained and fine-tuned speaker encoder and Tacotron 2 synthesizer, along with a HiFi-GAN neural vocoder model that involves a transfer-learning approach to synthesize child speech. A publicly available multi-speaker child speech corpus was cleaned and a pre-processed subset of 1-hour data was utilized for training and fine-tuning the proposed TTS model. The quality of the synthesized child’s speech was evaluated using the MOSNet score. The best quality of synthesized child speech achieved an average MOSNet score of 2.8 for the fine-tuned HiFi-GAN vocoder. The proposed TTS model could generate synthesized child speech on adoption of the transfer-learning approach.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Speech synthesis; text-to-speech; multi-speaker; Tacotron 2; HiFi-GAN; MOSNet
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science B Philosophy. Psychology. Religion > Psychology > Child psychology P Language and Literature > P Philology. Linguistics > Computational linguistics. Natural language processing Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Tamara Malone
Date Deposited:	11 Mar 2023 13:13
Last Modified:	11 Mar 2023 13:13
URI:	https://norma.ncirl.ie/id/eprint/6308

Actions (login required)

View Item