WaveNet: Dilated Receptive Field
Intermediate
The problem, from first principles
WaveNet (van den Oord et al., 2016) models raw audio one sample at a time with stacked causal convolutions whose dilation doubles each layer. The receptive field grows exponentially with depth, so a few layers cover thousands of samples.
From theory to code
Implement wavenet_receptive_field(dilations, kernel), the number of past samples an output sees.
Constraints
- Causal convolutions use only past samples.
Hints
Hint 1
Start at one sample, then add (kernel minus one) times the dilation for each layer.
Keywords
research-paperssequence-modelsgenerative-audioconvolution
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.