search

philipperemy / deep-speaker

Deep Speaker: an End-to-End Neural Speaker Embedding System.
MIT License
905 stars 241 forks source link

Memory usage #70

Closed lnguyen closed 3 years ago

lnguyen commented 3 years ago

Is there a way not to use so much memory if you add speakers?

philipperemy commented 3 years ago

You mean for training? I know the memory usage is not optimized for the softmax pre-training part (Somehow Keras needs double of the memory to work...). But the triplet training should be efficient in memory.

lnguyen commented 3 years ago

got it, thanks!