Vector-Quantized Contrastive Predictive Coding (VQ-CPC)
for Acoustic Unit Discovery and Voice Conversion

^{Fig 1: VQ-CPC model architecture.}

All audio samples are generated using the scripts and pretrained weights at https://github.com/bshall/VectorQuantizedCPC.

For samples from the VQ-VAE model see https://bshall.github.io/ZeroSpeech/.

English samples

Speaker - V001

Speaker - V002

Indonesian samples

Speaker - V001