Neural Speech Synthesis For Resource-Scarce Languages

The following speech samples relate to a paper submitted for the South African Forum for Artificial Intelligence Research 2019

The paper's bibtex citation is as follows:

@inproceedings{louw2019neuralspeech,
author = {Louw, Johannes A},
title = {Neural speech synthesis for resource-scarce languages},
booktitle = {Proceedings of the South African Forum for Artificial Intelligence Research},
year = {2019},
editor = {Etienne Barnard and Marelie Davel},
pages = {103--116},
month = {December},
address = {Cape Town, South Africa}
}

Systems are as follows (as described in the paper)

System name System description
DNN LPCNet Stacked bottleneck feed-forward neural network based voice with LPCNet vocoder
DNN World Stacked bottleneck feed-forward neural network based voice with World vocoder
HMM LPCNet HTS HMM-based voice with LPCNet vocoder
HMM World HTS HMM-based voice with World vocoder

Listening test stimuli

Utterance Reference DNN LPCNet DNN World HMM LPCNet HMM World
"Telkens moet hy ook die gevolge van sy dade dra."
"Uitslaap in die veld en eie kamp met eie geriewe opstel."
"Die kom handig te pas om rigting aan te dui."
"Hulle tradisionele geloof en waardes is geskud."
"Jou finansiƫle raadgewer kan daarmee help."
"Alles vorm 'n eenheid met die eenvoud van die omgewing."
"Gangbaar, maar nie uitdaging waarna ernstige spelers soek nie."
"Ek kan dus nou ook u aandag daarop vestig."
"Selfs die geskubde vel oor die oog het behoue gebly."
"Selfs sy eie ouers het teen hom gedraai."