基于线性预测模型的氦语音增强算法研究
Helium speech enhancement based on linear predictive coding
-
摘要: 潜水员在水下工作时,由于生理方面的原因,需要以氦氧混合气体作为呼吸气体,由于气体结构的变化出现了氦语音现象,使得语音发生畸变,降低了清晰度。通过介绍基于线性预测模型的氦语音增强算法,将频域的线谱对(LSP)分析应用于氦语音增强中,由于线谱对系数与语音信号谱包络有紧密的联系,用线谱对参数(LSP)构成合成滤波器时更容易保证稳定性,所以提出了一种基于线谱对系数(LSP)的氦语音增强算法。通过实验将这种算法与基于线性预测LPC的增强算法进行了比较,实验结果表明,两种算法均能对氦语音进行矫正,并且新的算法能够对共振峰进行单独调节,在不影响清晰度的同时最大限度地保持了原有语音的细节,提高了可懂度。Abstract: When divers work in deep sea,they must breathe helium-oxygen mixture in stead of air.Since the vocal tract is adapted to produce speech in normal atmosphere,a changed compo-sition of breath mixture causes distortions in the speech.Line-spectrum pair(LSP) associated with speech spectrum leads to synthesis filter based on LSP will never degrade system steady-state performance.Moreover LSP are important parameters of vocal track model in speech coding.Their quantization is closely related to synthetic speech quality.An LSP-based algorithm for helium speech enhancement is proposed in this paper.Having analyzed the resulting data,we find the universal LSP coefficient in the new algorithm.By comparison with the LPC method,it is seen that the pro-posed algorithm can correct helium speech leading to significant enhancement of intelligibility.
下载: