Speech attribute classifier using support vector machine for speech packet loss concealment
Hsi-Sheng Hsu, Yu-Jui Hung, Zhong Hua Hsu, Jui‐Feng Yeh · 2012
This paper presents a speech attribute classification method for speech packet loss concealment on voice over internet protocol (VOIP). Besides short time stable characteristics of speech signal, the features of syllable and sub-syllable are also used to packet loss concealment in this paper. Vicinity information is essential to reconstruct the lost voice data. Considering of the hearing and semantic understanding, the syllable and sub-syllable information is important for speech packet loss concealment. However, the speech recognition with higher computational complexity is not able to provide the real time processing for voice communication. This paper uses speech attributes to classify speech categories that are used to reconstruct the loss data by trained models. The experimental results show that the performance in speech attribute classifier is effective and efficient enough to reconstruct speech signals.