about the ar loop?

[azraelkuan]

from the code:

cargan/cargan/core.py

Line 212 in 61051fa

signals[start:start + cargan.CHUNK_SIZE] += signal.squeeze()

for each chunk output samples, it will be add to signals,
but in the for loop,

cargan/cargan/core.py

Line 207 in 61051fa

for i in range(0, features.shape[2] - feat_chunk + 1, feat_hop):

we have the feat_hop,,
for my understand, it will cumsum on the signals, but we only need the first feat_hop * hop_size samples, right?

[azraelkuan]

there is boundary mismatch between different chunks, how to solve this?

[maxrmorrison]

The code is written to allow the hop size and chunk size to be different. But if you check their values, they are the same. I experimented with a smaller hop size for a while, but found hop size == chunk size to be best.

As mentioned in the conclusion of our paper, we don't yet know how to solve the boundary artifacts. However, our experiments indicate that these artifacts are less severe than the periodicity artifacts.