Abstract:
"
Voice Conversion is a subset of Voice Transformation technique that specializes in
changing one speaker identity to another. Voice Conversion usually depends on
parallel data or non-parallel for training data. The trouble with data is that its hard-to find proper parallel data for voice conversion training To eliminate this models based
on non-parallel data was developed but even then non parallel data are hard to find as
well. Therefore in this project the author will be using a CycleGAN based approach
combined with a Mixup augmentation technique than can extend the data on non parallel speech corpus"
