基于卷积神经网络的鲁棒性基音检测方法

在语音信号中, 基音是一个重要参数, 且有重要用途. 然而, 检测噪声环境中语音的基音却是一项难度较大的工作. 由于卷积神经网络(Convolutional neural network, CNN)具有平移不变性, 能够很好地刻画语谱图中的谐波结构, 因此我们提出使用CNN来完成这项工作. 具体地, 我们使用CNN来选取候选基音, 再用动态规划方法(Dynamic programming, DP)进行基音追踪, 生成连续的基音轮廓. 实验表明, 与其他方法相比, 本文的方法具有明显的性能优势, 并且 对新的说话人和噪声有很好的泛化性能, 具有更好的鲁棒性.

Convolutional Neural Network for Robust Pitch Determination

Pitch is an important characteristic of speech and is useful for many applications. However, pitch determination in noisy conditions is difficult. Because shift-invariant property of convolutional neural network (CNN) is suitable to model spectral feature for pitch detection, we propose a supervised learning algorithm to estimate pitch using CNN. Specifically, we use CNN for pitch candidate selection, and dynamic programming (DP) for pitch tracking. Our experimental results show that the proposed method can obtain accurate pitch estimation and that it has a good generalization ability in terms of new speakers and noisy conditions.