__init__() got an unexpected keyword argument 'max_iter'

[wangjifly1]

when i run a test as follows:
python -m unittest test.audio_to_image_test
an error happens:

Traceback (most recent call last):
File "/Users/wangji/riffusion-main/test/audio_to_image_test.py", line 42, in test_stereo
self.helper_test_with_params(params)
File "/Users/wangji/riffusion-main/test/audio_to_image_test.py", line 60, in helper_test_with_params
audio_to_image(audio=str(audio_path), image=str(image_path), **params)
File "/Users/wangji/riffusion-main/riffusion/cli.py", line 54, in audio_to_image
converter = SpectrogramImageConverter(params=params, device=device)
File "/Users/wangji/riffusion-main/riffusion/spectrogram_image_converter.py", line 21, in init
self.converter = SpectrogramConverter(params=params, device=device)
File "/Users/wangji/riffusion-main/riffusion/spectrogram_converter.py", line 87, in init
self.inverse_mel_scaler = torchaudio.transforms.InverseMelScale(
TypeError: init() got an unexpected keyword argument 'max_iter'

Many thanks for help！！

[wangjifly1]

I have solved this problem :
pip install --upgrade torch==2.0.1
pip install --upgrade torchaudio==2.0.1
pip install --upgrade torchvision==0.15.1

[Cryptospy]

same, TypeError: init() got an unexpected keyword argument 'max_iter'

[J4BEZ]

@Cryptospy It is due to the stable version of torchaudio has been updated to 2.1.2,

InverseMelScale class in torchaudio 2.0.1 has the keyword argument named 'max_iter'
but the version of 2.1.2 doesn't.

You can solve the problem easily with the solution that @wangjifly1 suggested for us.
:: downgrade version of torchaudio

if you would be available for a second for me,
You may check it out in more detail through the PR that I wrote [ #165 ]

Hope you have a peaceful day and happy new year✨

[Cryptospy]

@J4BEZ This worked for me:

pip uninstall torch torchaudio torchvision

pip install torch==2.0.1+cu118 torchaudio==2.0.2+cu118 torchvision==0.15.2+cu118 -f https://download.pytorch.org/whl/cu118/torch_stable.html

[Cryptospy]

@J4BEZ happy new year to you too mate

[J4BEZ]

@Cryptospy Happy to hear that😆😆
Hope you have a joyful weekend and happy new year🙌🙌

[winterer]

If downgrading torchaudio is not an option, the following code modification works for me (removal of unused parameters due to API changes in torchaudio 2.1.x):

diff --git a/riffusion/spectrogram_converter.py b/riffusion/spectrogram_converter.py
index 9fbfc65..2beb605 100644
--- a/riffusion/spectrogram_converter.py
+++ b/riffusion/spectrogram_converter.py
@@ -90,10 +90,6 @@ class SpectrogramConverter:
             sample_rate=params.sample_rate,
             f_min=params.min_frequency,
             f_max=params.max_frequency,
-            max_iter=params.max_mel_iters,
-            tolerance_loss=1e-5,
-            tolerance_change=1e-8,
-            sgdargs=None,
             norm=params.mel_scale_norm,
             mel_scale=params.mel_scale_type,
         ).to(self.device)