Automatic segmentation and labeling for continuous number recognition
S. A. R. Al-Haddad, Salina Abdul Samad, Aini Hussein, Khairul Anuar Ishak, Akram Abdul Azid, R. Ghaffar, Dzati Athiar Ramli, M Zainal, Muhammad Abdullah · International Conference on Signal Processing · 2006
This study is focused on continuous number speech recognition with the intention to distinguish speech and non-speech segments and segment it as one digit. This study proposes an algorithm for automatic segmentation of male and female voiced speech. The calculations of log energy and zero rate crossing are used to process speech samples to accomplish the segmentation. The thresholds are set based on the maximum likelihood for accurate labeling parts of speech (POS). The algorithms manage to get 95% correct segmentation for male speakers and 72.5% from female speakers.