Search papers, labs, and topics across Lattice.
1
0
3
2
Jointly training a speech encoder and language model on mel-spectrograms not only boosts zero-shot speech translation, but also fixes annoying speech synthesis quirks like endless silences.
