【摘要】：提出了計(jì)算語(yǔ)音信號(hào)線譜對(duì)(LSP)參數(shù)的高效算法NRSPF。首先利用牛頓-拉夫森函數(shù)及斯蒂芬森加速求高階非線性方程的一個(gè)實(shí)根,再使用多項(xiàng)式綜合除法降階,最后采用費(fèi)拉里算法求其余的根,即得LSP參數(shù)。通過TI-DSP平臺(tái)的實(shí)例研究表明,NRSPF算法與APF算法相比,迭代次數(shù)減少、收斂速度加快,計(jì)算量小,并且在精度提高10倍、100倍和1000倍情況下,APF算法可能出現(xiàn)被零除錯(cuò)誤和死循環(huán),而NRSPF算法不僅避免了該錯(cuò)誤,而且迭代次數(shù)增加很少,收斂速度仍然很快,得到更精確的結(jié)果。本文提出的算法高效、可靠、實(shí)時(shí)性強(qiáng),可應(yīng)用于超低碼率語(yǔ)音實(shí)時(shí)通信系統(tǒng)、語(yǔ)音編解碼器等。
[Abstract]:An efficient algorithm, NRSPF., for calculating the (LSP) parameters of speech signal line spectrum is proposed. First, we use Newton-Raphson function and Stephenson's acceleration to find a real root of higher order nonlinear equation, then reduce the order by polynomial synthesis division method. Finally, we use Ferrari algorithm to find the other roots, that is, LSP parameter. A case study on the TI-DSP platform shows that compared with the APF algorithm, the TI-DSP algorithm has fewer iterations, faster convergence speed and less computation, and it may have zero division errors and dead cycles in the case of 10 times higher accuracy and 1000 times higher accuracy. The NRSPF algorithm not only avoids this error, but also increases the number of iterations very little, and the convergence rate is still fast, and the more accurate results are obtained. The algorithm proposed in this paper is efficient, reliable and real-time. It can be used in ultra-low bit-rate speech real-time communication system, speech codec and so on.
【作者單位】： 蘇州大學(xué)電子信息學(xué)院;
【基金】：國(guó)家自然科學(xué)基金(61271360,61201213) 蘇州市應(yīng)用基礎(chǔ)研究計(jì)劃(SYG201230)資助
【分類號(hào)】：TN912.3
，

本文編號(hào)：2219227