Speech Enhancement Based on Frequency Domain ALE with Adaptive De-Correlation Parameters
Isao Nakanishi, Hironori Namba, Shigang Li · International Journal of Computer Theory and Engineering · 2013
We have proposed the speech enhancement method using the frequency domain adaptive line enhancer, which enables to independently set de-correlation parameters at frequency bins. In the conventional method, we divided the frequency band into four domains and set a de-correlation parameter at each domain. However, the spectral distribution of voices varies depending on gender and/or age differences. In this paper, we propose to adaptively control the de-correlation parameter according to the signal-to-noise ratio at each domain. Moreover, we propose to adjust the frequency width of each domain to a detected pitch (fundamental frequency). Their effectiveness is evaluated in simulations.