search

WeidiXie / VGG-Speaker-Recognition

Utterance-level Aggregation For Speaker Recognition In The Wild
363 stars 97 forks source link

about the default num_class in the function vggvox_resnet2d_icassp() #56

Closed foofybuster closed 4 years ago

foofybuster commented 4 years ago

I saw the num_class param not inconsistent with detail in the paper. Should it be 5994 classes? And where can I find more data to combine with Voxceleb2(dev) and get a total of 8361 classes? Anyway, thank you so much for this implementation! :)

WeidiXie commented 4 years ago

can you point me where did you see the 8361 number ?

foofybuster commented 4 years ago

Sure! image it's in file model.py.

**8631 classes, I mistook. Sorry.

WeidiXie commented 4 years ago

that's the default number, I simply used the previous code I have, the 8631 was adapted from the VGGFace2 paper and code.

In Voxceleb2, we only have 5994 classes.

foofybuster commented 4 years ago

Thank you so much!