【摘要】 为了增强多带激励(MBE)声码器基音估计性能的鲁棒性,提出了一种可适用于低信噪比语音信号的改进双路径基音跟踪算法.采用全新构造的差值不等式作为约束方程,其差值门限的取值在基音跟踪过程中能够根据基音周期长短的统计特征自动更新.实验结果显示:在SNR为-5dB的高斯白噪声干扰的情况下,基音估计的严重错误概率的性能改善平均达到70%.与传统算法相比,该算法对不同讲话者和不同程度高斯白噪声干扰均具有较强的适应能力,尤其在噪声严重的情况下该算法对基音估计的准确性得到明显改善,从而使合成语音具有较好的可懂度和自然度.更多还原

【Abstract】 An improved two-path pitch tracking algorithm applicable to noisy speech signals at low signal to noise ratio(SNR) levels was proposed to enhance robustness of pitch estimation performance for multiband excitation(MBE) vocoder.The algorithm employed a new constructed difference inequation as the constraint equation,and its difference threshold value could be updated automatically during the pitch tracking process according to the statistical feature of the pitch period length.The experimental results show that the average gross error rate improvement achieves 70% at-5 dB SNR level in white Gaussian noise condition.The algorithm outperforms the conventional algorithm,displaying better adaptability to different speakers and different white Gaussian noise conditions,improving the pitch estimation accuracy obviously especially in serious noise conditions,thus facilitating more intelligible and more natural synthesized speech.更多还原