| START Conference Manager |
category: Poster Session: 5 December Session P1: Oceanian Languages Poster Session
Abstract: We have made a synthetic male te reo Māori voice, which runs on MaryTTS and synthesises speech from any Māori text. The voice was created from recordings of 1030 sentences, chosen to ensure full diphone coverage, from the story Ngā Mahi a Ngā Tūpuna. The recordings were made in a soundproof booth and condensed to two hours of continuous speech. Phonetic labelling was first determined automatically using a model obtained from Montreal Forced Aligner. The labelled data, along with a 10,000-word Māori lexicon with phonetic transcription and stress mark up, is passed into MaryTTS to create a synthetic voice. Resume: Kua hangaia e mātou he reo rorohiko, he reo tāne. Ka noho tēnei reo rorohiko ki roto i a te pūmanawa MaryTTS. Ka whakaurua e tangata he tuhinga, ā, ka hurihia he reo, arā he kōnae tangi. Kua hopukia tēnei ki roto i tētahi wharau karo tangi. I ahu mai ngā kōrero i te pukapuka Ngā Mahi a Ngā Tupuna, ā, kua wehewehe ngā kōrero kia rerenga kōrero poto (1030). Ka āta whiriwhiria ētahi anō kia mau te katoa o ngā tangi oro o te reo Māori. Kia hangai atu ngā oro tuhi ki ngā oro tangi, ka whakamahia te Montreal Forced Aligner, nā te tangata anō i atā tirotiro te tika o ngā hononga. Ka tukua ēnei raraunga ki a MaryTTS, ka tukua hoki he papakupu whakaahua (10,000 + ngā kupu).
[Paper (PDF)]
START Conference Manager (V2.61.0 - Rev. 5964)