Development of a Tri-Pitch and Tri-Tone Utau VoiceBank: Hanakoloid

This research study describes the development of a CVVC (Consonant-Vowel-Consonant-Vowel) voice bank for UTAU, a popular vocal synthesis software. This paper states that the creative process involved a deep analysis of Tamil in order to represent its peculiar phonetics. To synthesize the voice clearly and smoothly using a variety of vocal tract changes, with accuracy, it is essential to examine in detail the phonemes in Tamil and establish an accurate list known as recording list (reclist). Voicebank is versatile, as it has been recorded in three pitches in each tone (Power, Soft & Cute) to enhance vocal expression. For Tamil and other South Indian languages special care was that transitions between phonemes were not abrupt for the output to sound natural, as well it was made sure minimal accent while in Hindi. This final voicebank, weighing in at nearly 1GB has been extensively tested and finetuned to provide exceptional and human-like vocal performances. This project is a major milestone in the world of vocal synthesis, particularly for Tamil and other Dravidian languages expanding UTAU's linguistic capabilities.