Similarity-and-Independence-Aware Beamformer: Method for Target Source Extraction using Magnitude Spectrogram as Reference
share
This study presents a novel method called the similarity-and-independence-aware beamformer (SIBF) for source extraction. The SIBF can extract the target signal using its rough magnitude spectrogram as the reference signal. The advantage of SIBF lies in that it can obtain an accurate target signal, compared to the spectrogram generated by the target-enhancing methods, such as the speech enhancement based on deep neural networks (DNNs). To realize such extraction, we extend the framework of the deflationary independent component analysis, by considering the similarity between the reference and extracted target, as well as the mutual independence among all potential sources. To solve this extraction problem by the maximum-likelihood estimation, we introduce two types of source models that can reflect the similarity. Using the CHiME3 dataset, the experimental results show that the SIBF can extract the target signal more accurate than the reference generated by the DNN.
READ FULL TEXT
