Input format about wavenet-enhancement HOT 5 CLOSED

The shape of all inputs is width * num_of_tokens, where widths are different for clean, noisy, and target.
Clean input and target are quantized. Noisy input is raw audio samples.
You need to create data matrix with appropriate shape and alignment. I used HDF5 just because my data matrix is created using Matlab.
I should mention all of these in Readme later.

from wavenet-enhancement.

First of all thanks for replay. have to clarify the above comments,
what you meant by the 'num_of_tokens', is that the output space size (ie 256)? or simply the strength of my data set .

from wavenet-enhancement.

num_of_tokens is the size of the data set, similar to "batch_size".
channel is the output space size you are referring to.

from wavenet-enhancement.

could you tell me how to create the input data?

from wavenet-enhancement.

The clean input is the quantized audio ranging from 0 to 255.
The noisy input is the corresponding noisy version in raw audio samples.

from wavenet-enhancement.