Only works for speakers in the training dataset?

auspicious3000 / SpeechSplit

Unsupervised Speech Decomposition Via Triple Information Bottleneck

http://arxiv.org/abs/2004.11284

MIT License

636 stars 92 forks source link

Only works for speakers in the training dataset? #61

Closed zhouyong64 closed 2 years ago

zhouyong64 commented 2 years ago

After reading the experiement part of your paper, it seems to me that SpeechSplit only works for speakers in the training dataset. Is this correct? How is its performance for speakers unseen during the training stage?