Harmonic plus Noise Model Combined with Support Vector Machine for Speech Packet Loss Concealment

• Main Authors: Zhong-Hua Hsu, 許中華
• Other Authors: Jui-Feng Yeh
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/89671937775131709919

碩士 === 國立嘉義大學 === 資訊工程學系研究所 === 99 === This paper presents an approach for speech packet loss concealment that based on Harmonic plus Noise Model (HNM) for voice over internet protocol (VOIP). Traditional packet loss concealment technique is based on short time stable characteristics for speech, by packet loss in the vicinity of the voice features to reconstruct the lost voice data. Once the packet loss is too long, there are no short-term stability characteristics, so the performance of traditional packet loss concealment technique will drop in this case. For a longer length of packet loss in case there are good voice qualities, this paper training the model to describe the voice changes over time. The Spectral Distance (SD), Signal-to-Noise Ratio (SNR) and the Mean Opinion Score (MOS) are used to evaluate the proposed method on different packet loss length. The experimental results show that the quality of the reconstructed speech signal of the proposed method outperforms that the traditional methods.