Update variable names in wavernn model #797
Conversation
Codecov Report
@@ Coverage Diff @@
## master #797 +/- ##
==========================================
+ Coverage 89.82% 89.84% +0.01%
==========================================
Files 34 34
Lines 2654 2648 -6
==========================================
- Hits 2384 2379 -5
+ Misses 270 269 -1
Continue to review full report at Codecov.
|
test/test_models.py
Outdated
|
|
||
| def test_mol(self): | ||
| """Validate the output dimensions of a _WaveRNN model in mol mode. | ||
| """Validate the output dimensions of a _WaveRNN model in mol loss. |
There was a problem hiding this comment.
This test has been removed.
torchaudio/models/_wavernn.py
Outdated
| if self.mode == 'waveform': | ||
| self.n_classes = 2 ** n_bits | ||
| elif self.mode == 'mol': | ||
| if self.loss == 'crossentropy': | ||
| self.n_classes = n_classes | ||
| elif self.loss == 'mol': | ||
| self.n_classes = 30 | ||
| else: | ||
| raise ValueError(f"Expected mode: `waveform` or `mol`, but found {self.mode}") | ||
| raise ValueError(f"Expected loss: `crossentropy` or `mol`, but found {self.loss}") |
There was a problem hiding this comment.
We discussed offline that we would remove the "loss/mode" parameter for WaveRNN. Let's do it as part of this PR.
There was a problem hiding this comment.
Yes, I have removed it. I also removed the sample_rate parameter since it is not needed.
torchaudio/models/_wavernn.py
Outdated
| n_hidden: the number of hidden dimensions (default=128) | ||
| n_output: the number of output dimensions (default=128) | ||
| n_hidden_resblock: the number of hidden dimensions of resblock (default=128) | ||
| n_output_melresnet: the number of output dimensions of melresnet (default=128) |
There was a problem hiding this comment.
These are clear as they are for a MelResBlock. Let's keep n_hidden and n_output here, but add "of resblock" and "of melresnet" in the description as you have done here.
There was a problem hiding this comment.
I agree, I kept them as before and changed the description.
Update audio preprocessing tutorial
2 participants
This is to update the variable names in WaveRNN model.
Related to #735