Towards The Development Of Accent Conversion Model For (L1)Bengali Speaker Using Cycle Consistent Adversarial Network (Cyclegan)

The goal of foreign accent conversion (FAC) is to create a new voice with the voice identity of a given second-language (L2) speaker but with a native (L1) accent. The main motivation for this work is to solve the problem of Indians understanding foreigners’ accents and foreigners understanding Indian accents. We propose a non-parallel accent conversion (AC) method that can learn mapping from source to target without the use of parallel speech data. The proposed method is particularly advantageous in that it does not require any additional data or an alignment procedure. A cycle-consistent adversarial network is used in our proposed method. This enables the identification of an optimal pseudo pair. from nonparallel dataset. We applied our method on a speech data contained only American English and English speech recorded by Bengali speaker. As per our knowledge this is the first attempt to convert the accent native (L1 American English speaker) and nonnative (L1 Bengali speaker). We trained our model under disadvantageous condition (non-parallel data). Result shows that this approach will be useful in accent conversion field.