Unofficial implementation result of Google AI's VoiceFilter system

Implementation GitHub repo link: mindslab-ai/voicefilter

Original Paper: arXiv

In summary, VoiceFilter isolates the voice of a desired speaker from noisy input, given the prior of that speaker. This helps reducing word error rate(WER) of speech recognition system.

Input 1: Mixed audio Input 2: Reference audio for d-vector Result: Estimated audio Ground truth

All details are shown in GitHub repository's README. Thank you!

Implementation author: Seungwon Park at MINDsLab