当前位置: X-MOL 学术Analog Integr. Circ. Signal Process. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Improved phase aware speech enhancement using bio-inspired and ANN techniques
Analog Integrated Circuits and Signal Processing ( IF 1.4 ) Pub Date : 2019-12-07 , DOI: 10.1007/s10470-019-01566-z
Tusar Kanti Dash , Sandeep Singh Solanki , Ganapati Panda

The phase modification of noisy speech signal plays a crucial role in speech enhancement (SE). In the recent past, many speech denoising algorithms have been proposed using the modification of phase information which depends on the scaling factor computed from the noise level. The performance measures of SE is significantly affected by this scaling factor and noise level estimation. However, in these algorithms, the parameters are not optimally tuned for the different noise conditions and also in some cases, the background noise is presumed to be stationary. Further, no earlier attempt has been made to obtain adaptive models which can establish the relationship between noise levels and scaling factor. Being motivated by these observations an attempt has been made in this paper to develop a neural network based model which is capable of properly estimating this scaling factor from the noise level. In the current work, a popular and efficient bio-inspired technique known as firefly algorithm is employed to determine the best possible scaling factor for each noise level. In addition, a relationship is established between noise level and scaling factor using trigonometric functional expansion based artificial neural network. An effective nonstationary noise estimation strategy is also incorporated in the proposed algorithm. Simulation-based experiments are performed to evaluate the effectiveness of the proposed SE algorithm and compared with other six standard SE algorithms using standard database. The analysis of the simulation results demonstrates that the proposed method outperforms the others in terms of both subjective and objective evaluation measures.

中文翻译:

使用生物启发和ANN技术改进相位感知语音增强

噪声语音信号的相位修改在语音增强(SE)中起着至关重要的作用。在最近的过去中,已经提出了利用相位信息的修改来提出许多语音去噪算法,该相位信息取决于根据噪声水平计算出的比例因子。SE的性能指标受此缩放因子和噪声水平估计的影响很大。但是,在这些算法中,对于不同的噪声条件,未对参数进行最佳调整,而且在某些情况下,背景噪声被认为是固定的。此外,尚未进行任何尝试来获得可以建立噪声水平和缩放因子之间的关系的自适应模型。受这些观察结果的激励,本文尝试开发一种基于神经网络的模型,该模型能够根据噪声水平正确估计此比例因子。在当前的工作中,一种流行且有效的生物启发技术被称为萤火虫算法,用于确定每种噪声水平的最佳可能缩放因子。另外,使用基于三角函数扩展的人工神经网络在噪声水平和比例因子之间建立了关系。一种有效的非平稳噪声估计策略也被纳入提出的算法中。进行基于仿真的实验以评估所提出的SE算法的有效性,并与使用标准数据库的其他六种标准SE算法进行比较。
更新日期:2020-01-04
down
wechat
bug