论文部分内容阅读
针对极低速率语音压缩编码中比特资源有限,量化精度严重不足的问题,该文提出了一种新的编码策略——减少量化传输的内容,提高重要内容的量化精度。语音经过低通滤波器将最不重要的3~4 kHz频谱滤掉,并相应的将采样率从8 kHz降低到6 kHz,同时保持每帧样点数不变。这样各个参数的联合帧数就减少为原来的3/4,在比特数不变的情况下,可以有效地提高量化精度。另外,对于线性预测系数(linear prediction coefficient,LPC)而言,由于语音谱从原来的0~4 kHz变为现在的0~3 kHz,LPC的预测阶数可以从10降低为8,参数维数降低,量化精度可以得到进一步提高。在此框架下,结合子带清浊音(band-pass voicing,BPVC)解码端恢复算法,实现了高质量极低速率150 b/s语音压缩编码算法。与现有的两种150 b/s算法相比,客观平均意见得分(mean opinion score,MOS)分别提高了0.051和0.067,同时LPC参数的谱失真分别降低了0.09和0.16,改进了合成语音质量,提高了可懂度。
Aiming at the problem of limited bit resources and insufficient quantization precision in very low rate speech compression coding, this paper proposes a new coding strategy - reducing the content of quantization transmission and improving the quantization precision of important content. The speech passes through the low-pass filter to filter out the least significant 3 to 4 kHz spectrum and correspondingly reduces the sampling rate from 8 kHz to 6 kHz while keeping the number of samples per frame constant. In this way, the joint frame number of each parameter is reduced to three-fourths of the original number, and the quantization precision can be effectively improved under the condition of the same number of bits. In addition, for the linear prediction coefficient (LPC), since the speech spectrum changes from the original 0 to 4 kHz to the current 0 to 3 kHz, the prediction order of the LPC can be reduced from 10 to 8, and the parameter dimension Reduce, the quantization precision can be further improved. In this framework, a high-quality and ultra-low-rate 150 b / s speech compression coding algorithm is implemented in combination with the band-pass voicing (BPVC) decoding end recovery algorithm. Compared with the two existing 150 b / s algorithms, the mean opinion score (MOS) increased by 0.051 and 0.067, respectively, while the spectral distortion of LPC parameters decreased by 0.09 and 0.16, respectively, which improved the quality of synthesized speech , Improve the intelligibility.