Discovering Interpretable Acoustic Features in Autoregressive Voice-Cloning TTS Models
Probes which acoustic features voice-cloning models actually use.
Explainer · on the wire
The explainer for this story is still on the wire.
The claim, the mechanism, and the figures land here once generated.
Source · arXiv