Speech Dataset Development for a Low-Resource Tibeto-Burman Tonal Language
Thiyam Susma Devi, Pradip K. Das · 2023
Manipuri is a low-resource, Tibeto-Burman Tonal Language spoken in the northeast part of India. Assimilation of tones for speech recognition of tonal languages is essential. Extensive research has been conducted on detecting the tone of tonal languages Chinese, Mandarin, Vietnamese and Thai and a few on Indian tonal languages such as Mizo, Boro and Punjabi nevertheless, Manipuri is mainly uncharted as no dataset is available to study the tones of Manipuri language. In-depth studies of tones of tonal languages require minimal pairs of words that differ just by pitch. To address this issue, this paper develops a speech corpus of isolated tonal contrast word pairs in Manipuri called ManiTo with two variants: ManiToSpc and ManiToScl, representing the spectrogram and the scalogram of the audio files to adopt the machine learning and deep learning techniques effectively.