NASAM 2.0: Cleft-Palate Speech Assessment Application

Nattapong Kurpukdee, Kwanchiva Thangthai, Vataya Chunwijitra, Patcharika Chootrakool, Sawit Kasuriya

Abstract: This paper describes the incremental development of Naso-ariticulometer (NASAM) equipment for assessing children with cleft lip and palate pronunciation. The NASAM collects patients’ words and sentences, predicts phoneme sequences, and delivers pronunciation and nasality scores. This program faces a number of issues, including identifying children’s speech, applying automatic phoneme recognition to exact speech evaluation, and developing the application’s UI/UX for both children and language and speech assessors. This study focuses on enhancing acoustic modeling for Thai children aged 6 to 13 years old in the phoneme recognition module. We address the issue of determining an acceptable acoustic characteristic for children’s speech and compare acoustic model training methods. The phone error rate (PER) using collected speech from 75 cleft children and 104 normal children is used to test the acoustic models. The experimental results show that the vocal tract length normalization method provides the best PER in normal children, resulting in a 19.17 percent relative error reduction from the baseline, while Mel-filter bank and pitch performs well in cleft children, resulting in a 4.61 percent relative error reduction from the baseline. Additionally, we conducted a satisfaction poll during the function test and discovered that the most pleasing response was 42 percent.