Autoencoder Based Architecture For Fast & Real Time Audio Style Transfer

Abstract

Recently, there has been great interest in the field of audio style transfer, where a stylized audio is generated by imposing the style of a reference audio on the content of a target audio. We improve on the current approaches which use neural networks to extract the content and the style of the audio signal and propose a new autoencoder based architecture for the task. This network generates a stylized audio for a content audio in a single forward pass. The proposed network architecture proves to be advantageous over the quality of audio produced and the time taken to train the network. The network is experimented on speech signals to confirm the validity of our proposal.

A Style Transfer Approach to Source Separation

“…In the case of style transfer problems for audio data, style and content are ambiguous [18]. Consequently, style transfer approaches for audio have largely resulted in efforts that match the timbres of the source and target domains (for example, opera sounds have been matched to sound like cats licking milk) [18,19]. In such cases, it is difficult to define what a good network output is supposed to sound like.…”