In this paper, we propose an anomization method that obfsucate the content and msask the speaker identity, while preserving the acoustic scene. s
In this paper, we present our data augmentation algorithm, tailored for singing voice extraction
We investigated different input representation to detect music structure boundaries.
We constructed the largest dataset of audio and lyrics aligned, using singing voice detection with deep learning.
Second paper on the DALI dataset, presenting 2nd version. An human evaluation of the produced alignment is also presented.