Audio samples for Voice Conversion Challenge

Description

We present the conversions generated by our method to the 2018 voice converison challenge. Samples are provided for both Hub and Spoke tasks. The model was trained using 5 minutes of audio for each speaker. Each table cell links to the complete set of conversions for the matching source and target speakers.

Hub taks

TF1 TF2 TM1 TM2
SF1 SF1 → TF1 SF1 → TF2 SF1 → TM1 SF1 → TM2
SF2 SF2 → TF1 SF2 → TF2 SF2 → TM1 SF2 → TM2
SM1 SM1 → TF1 SM1 → TF2 SM1 → TM1 SM1 → TM2
SM2 SM2 → TF1 SM2 → TF2 SM2 → TM1 SM2 → TM2

Spoke taks

TF1 TF2 TM1 TM2
SF3 SF3 → TF1 SF3 → TF2 SF3 → TM1 SF3 → TM2
SF4 SF4 → TF1 SF4 → TF2 SF4 → TM1 SF4 → TM2
SM3 SM3 → TF1 SM3 → TF2 SM3 → TM1 SM3 → TM2
SM4 SM4 → TF1 SM4 → TF2 SM4 → TM1 SM4 → TM2