model loading issue

[008karan]

I am using pre-trained model to get the embedding but getting below error:

RuntimeError                              Traceback (most recent call last)
<ipython-input-5-4bcf886e7e7f> in <module>
      2 import torchaudio
      3 
----> 4 wav2mel = torch.jit.load("wav2mel.pt")
      5 dvector = torch.jit.load("dvector.pt").eval()
      6 

~/anaconda3/lib/python3.8/site-packages/torch/jit/_serialization.py in load(f, map_location, _extra_files)
    159     cu = torch._C.CompilationUnit()
    160     if isinstance(f, str) or isinstance(f, pathlib.Path):
--> 161         cpp_module = torch._C.import_ir_module(cu, str(f), map_location, _extra_files)
    162     else:
    163         cpp_module = torch._C.import_ir_module_from_buffer(

RuntimeError: 
Class Namespace cannot be used as a value:
Serialized   File "code/__torch__/torchaudio/sox_effects/sox_effects.py", line 5
    effects: List[List[str]],
    channels_first: bool=True) -> Tuple[Tensor, int]:
  in_signal = __torch__.torch.classes.torchaudio.TensorSignal.__new__(__torch__.torch.classes.torchaudio.TensorSignal)
                                                                      ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
  _0 = (in_signal).__init__(tensor, sample_rate, channels_first, )
  out_signal = ops.torchaudio.sox_effects_apply_effects_tensor(in_signal, effects)
'apply_effects_tensor' is being compiled since it was called from 'SoxEffects.forward'
Serialized   File "code/__torch__/data/wav2mel.py", line 29
    wav_tensor: Tensor,
    sample_rate: int) -> Tensor:
    _0 = __torch__.torchaudio.sox_effects.sox_effects.apply_effects_tensor
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
    _1 = _0(wav_tensor, sample_rate, self.effects, True, )
    wav_tensor1, _2, = _1

[yistLin]

I myself also encountered this after upgrading torch to 1.8.0 and torchaudio to 0.8.0. I will try my best to fix this ASAP!

[ADD-eNavarro]

Looking forward for this to be fixed, since I find myself in the same situation.
Thanks for your great work, by the way!

[yistLin]

I just realized that the default backend has been changed from sox to sox_io in torchaudio 0.8.0. I'm not very sure whether this causes the problem, but ever since I recompiled the Wav2Mel module, it works just fine. Plan to upload the new module tomorrow.

[yistLin]

The newly compiled modules have been uploaded with the new release.

[ADD-eNavarro]

Great, thank you! Eager to try them!

[ADD-eNavarro]

Still didn't work for me:

Message=
Unknown builtin op: torchaudio::sox_effects_apply_effects_tensor.
Could not find any similar ops to torchaudio::sox_effects_apply_effects_tensor. This op may not exist or may not be currently supported in TorchScript.
:
File "/home/yist/.pyenv/versions/3.8.5/lib/python3.8/site-packages/torchaudio/sox_effects/sox_effects.py", line 151
>>> assert sample_rate == 8000
"""
return torch.ops.torchaudio.sox_effects_apply_effects_tensor(
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
tensor, sample_rate, effects, channels_first)
Serialized File "code/torch/torchaudio/sox_effects/sox_effects.py", line 5
effects: List[List[str]],
channels_first: bool=True) -> Tuple[Tensor, int]:
_0, _1 = ops.torchaudio.sox_effects_apply_effects_tensor(tensor, sample_rate, effects, channels_first)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
return (_0, _1)
'apply_effects_tensor' is being compiled since it was called from 'SoxEffects.forward'
Serialized File "code/torch/data/wav2mel.py", line 31
wav_tensor: Tensor,
sample_rate: int) -> Tensor:
_0 = torch.torchaudio.sox_effects.sox_effects.apply_effects_tensor
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
_1 = _0(wav_tensor, sample_rate, self.effects, True, )
wav_tensor1, _2, = _1

Source=***
StackTrace:
File "***", line 11, in (Current frame)
wav2mel = torch.jit.load(r"yistLin dvector\wav2mel.pt") # Error: #5

[yistLin]

Did you import torchaudio before loading the wav2mel.pt module?

[ADD-eNavarro]

Yes, my code is as in the example:

import torch # This is 1.8.0
import torchaudio # This is 0.8.0

wav2mel = torch.jit.load(r"yistLin dvector\wav2mel.pt") # Error

[ADD-eNavarro]

Exactly those (sorry, updated the comment a bit too late)

[yistLin]

@ADD-eNavarro Can you try to compile the modules? You can simply run the preprocessing script and those modules will be generated and saved with the preprocessed data. Or you can do it manually like this:

# in the root directory of the repo
import torch
import torchaudio

from data.wav2mel import Wav2Mel

wav2mel = Wav2Mel()
wav2mel = torch.jit.script(wav2mel)
wav2mel.save("wav2mel.pt")

[ADD-eNavarro]

I get an error in:
wav2mel = torch.jit.script(wav2mel)

Message=
undefined value torch:
File "\AppData\Roaming\Python\Python37\site-packages\torchaudio_internal\module_utils.py", line 151
>>> assert sample_rate == 8000
"""
return torch.ops.torchaudio.sox_effects_apply_effects_tensor(
~~~~~ <--- HERE
tensor, sample_rate, effects, channels_first)
'apply_effects_tensor' is being compiled since it was called from 'SoxEffects.forward'
File "\yistLin_dVector\data\wav2mel.py", line 72
def forward(self, wav_tensor: torch.Tensor, sample_rate: int) -> torch.Tensor:
wav_tensor, _ = apply_effects_tensor(wav_tensor, sample_rate, self.effects)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
return wav_tensor

Source=\yistLin_dVector\Test.py
StackTrace:
File "\yistLin_dVector\Test.py", line 7, in (Current frame)
wav2mel = torch.jit.script(wav2mel)

[yistLin]

Got it, now I see some problems:

1. Do you have sox installed on your machine?
2. From those logs you posted, I guess you're using Windows. But I'm not 100% sure whether torchaudio's SoxEffects works on Windows as on Linux/Mac OS.

Reading torchaudio's docs, it says that the sox_io backend works on Linux/Mac OS only, but I didn't see any warnings like: SoxEffects does not work on Windows...

[ADD-eNavarro]

You're right, I'm working on Windows.
Sox is installed, sox_io won't install on Windows. That might be the cause indeed. I will look into it too, but it would be a pity to discard Windows users like that...