Bug fix in encodec.py - Githubissues

lucidrains / audiolm-pytorch

Implementation of AudioLM, a SOTA Language Modeling Approach to Audio Generation out of Google Research, in Pytorch

MIT License

2.33k stars 249 forks source link

Closed yang1fan2 closed 1 year ago

yang1fan2 commented 1 year ago

Found two bugs in encodec.py

the concatenation of encoded should be last dimension instead of dim=1 (it's num_quantizers).
if return_encoded is false, exception will be thrown at line emb, = unpack(emb, ps, '* n c'), since emb is None.

lucidrains commented 1 year ago

@yang1fan2 oh thanks!