The Chulalongkorn Corpus of Spoken Thai (CCOST)

Abstract

The Chulalongkorn Corpus of Spoken Thai (CCOST) is a phonetically annotated corpus of Standard Thai. The corpus comprises approximately 7 hours of interview-style spontaneous speech from 49 speakers (19 male, 30 female) ranging in age from 18 to 83 years old. Speakers represent diverse regional backgrounds across Thailand but were instructed to speak in Standard Thai. Each speaker also read a 206-item monosyllabic word list twice and a set of 25 sentences three times. The annotation pipeline combines automatic speech recognition (ASR) and forced alignment using CLARIN-D’s OCTRA and Munich Automatic Segmentation System (MAUS) tools with manual correction by phonetically trained native Thai speakers. Transcriptions include orthographic, word-level, syllable-level, and phone-level annotations including toneme labels. The corpus serves as a resource in the sociophonetic investigation of segmental and tonal variation in spontaneous and controlled speech, enabling examination of individual characteristics as well as group differences across age groups, genders, and regional backgrounds. Hand-corrected annotations will additionally serve to improve forced alignment accuracy for Standard Thai.

Resources

Details