研究并改进了矢量和激励线性预测（ｖｅｃｔｏｒｓｕｍｅｘｃｉｔｅｄｌｉｎｅａｒｐｒｅｄｉｃｔｉｏｎ，ＶＳＥＬＰ）语音编码器．介绍了８ｋｂｉｔ／ｓＶＳＥＬＰ编码方案的实现及在此基础上利用矢量量化ＬＰＣ参数、减少激励码本数和子块分裂等方法；改进的６．４ｋｂｉｔ／ｓ，４．８ｋｂｉｔ／ｓ及２．４ｋｂｉｔ／ｓＶＳＥＬＰ的方案和结果．试听结果表明，６．４ｋｂｉｔ／ｓ，４．８ｋｂｉｔ／ｓ方案的合成语音质量接近８ｋｂｉｔ／ｓ的，具有较高的可懂度和自然度，而２．４ｋｂｉｔ／ｓ的质量明显下降． Research and improve the vector and excitation linear prediction (vectorsumexcitedlinearprediction, VSELP) speech coder. The implementation of 8kbit / s VSELP coding scheme is introduced. Based on this, vector quantization LPC parameters are used to reduce the number of excitation codes and sub-block splitting. The improved 6.4kbit / s, 4.8kbit / s and 2.4kbit / s VSELP Programs and results. The audition results show that the synthesized speech quality of 6.4 kbit / s and 4.8 kbit / s solutions is close to 8 kbit / s, with a high degree of intelligibility and naturalness, while the quality of 2.4 kbit / s is obviously decreased.