An Initial study on Birdsong Re-synthesis Using Neural Vocoders

TL;DR

This study compares traditional and neural vocoders for birdsong resynthesis, revealing similar species discrimination but differences in perceived bird-like qualities, highlighting challenges in processing wildlife audio.

Contribution

First comparative analysis of birdsong resynthesis using traditional and neural vocoders, identifying strengths and limitations in bio-acoustic applications.

Findings

No difference in species discrimination across vocoders

WORLD vocoder samples rated higher for bird-like qualities

All vocoders faced pitch and voicing challenges

Abstract

Modern speech synthesis uses neural vocoders to model raw waveform samples directly. This increased versatility has expanded the scope of vocoders from speech to other domains, such as music. We address another interesting domain of bio-acoustics. We provide initial comparative analysis-resynthesis experiments of birdsong using traditional (WORLD) and two neural (WaveNet autoencoder, parallel WaveGAN) vocoders. Our subjective results indicate no difference in the three vocoders in terms of species discrimination (ABX test). Nonetheless, the WORLD vocoder samples were rated higher in terms of retaining bird-like qualities (MOS test). All vocoders faced issues with pitch and voicing. Our results indicate some of the challenges in processing low-quality wildlife audio data.