Search papers, labs, and topics across Lattice.
International Audio Laboratories Erlangen, Fraunhofer Institute for Integrated Circuits IIS
3
0
3
Vocos excels at magnitude modeling but falters on phase, revealing critical architectural limitations that could redefine vocoding approaches.
Room embeddings can now be reliably estimated from reverberant speech with a calibrated uncertainty score, enabling selective prediction from just one utterance.
Speaker identity is effectively isolated in acoustic teleportation codecs, but acoustic features leak into speech embeddings, challenging existing evaluation methods.
