论文部分内容阅读
本文提出的声码器将语音分成静音、清音、浊音和混合音四类．用自适应方法进行分频带清浊音判决和有声／无声判决，提高了分类算法的稳定性、准确性和灵活性，还保持了混合语音的音质，且无须对清浊音判决结果进行编码．对清音和浊音的频谱分别采用不同的ＬＳＰ量化表进行编码，从而用标量量化器替代了矢量量化器，降低了复杂度．声码器的码率最高２４ｋｂｐｓ，最低为１００ｂｐｓ，平均码率１４ｋｂｐｓ．实时软件系统的延迟时间约０３秒．用４０ＭＨｚＴＭＳ３２０Ｃ５０定点ＤＳＰ实现了解码与合成部分的实时处理，平均运算量为１１３ＭＩＰＳ．
The vocoder proposed in this paper divides speech into four categories: mute, unvoiced, voiced and mixed. Using adaptive method to make sub-band voiced / unvoiced decision and voiced / unvoiced decision, the stability, accuracy and flexibility of classification algorithm are improved, the quality of mixed voice is also maintained, and the result of voiced / unvoiced decision is not required to be encoded. The spectra of unvoiced speech and voiced speech are encoded by different LSP quantization tables, so that the vector quantizer is replaced by a scalar quantizer, which reduces the complexity. Vocoder code rate up to 2 4kbps, a minimum of 100bps, the average rate of 1 4kbps. Real-time software system delay time of about 0 3 seconds. With 40MHzTMS320C50 fixed-point DSP decoding and synthesis part of the real-time processing, the average computing capacity of 11 3MIPS.