CLI Inference - What am I doing wrong?

[noisyneil]

I've tried:

svc -m <model path> -c <config path> -o <output path> -fm crepe -d CPU <input path>

...but I get:

Error: No such option: -m

Meanwhile the GUI version works great.

[Grownz]

Try svc infer -m ...

[noisyneil]

Try svc infer -m ...

I did know that, but erm... forgot. 🥴

When I included infer, I got this:

Error: Invalid value for '-m' / '--model-path': Path '<path to model file>' does not exist.

I figured it might be spaces in the folder paths that was causing the issue so I surrounded them each in double quotes and it seemed to start ok, but then I got this:

Click me

[17:50:45] WARNING  [17:50:45]                                   warnings.py:109
                    /usr/local/Caskroom/miniforge/base/lib/pytho                
                    n3.10/site-packages/so_vits_svc_fork/modules                
                    /synthesizers.py:81: UserWarning: Unused                    
                    arguments: {'n_layers_q': 3,                                
                    'use_spectral_norm': False}                                 
                      warnings.warn(f"Unused arguments:                         
                    {kwargs}")                                                  
                                                                                
           INFO     [17:50:45] Decoder type: hifi-gan        synthesizers.py:100
[17:50:46] INFO     [17:50:46] Loaded checkpoint                    utils.py:261
                    '/Users/noisyneil/Desktop/F - Lana Del Rey                  
                    100k/G_100000.pth' (epoch 3508)                             
[17:50:47] INFO     [17:50:47] Chunk: Chunk(Speech: False, 61740.0)  core.py:283
           INFO     [17:50:47] Chunk: Chunk(Speech: True, 264600.0)  core.py:283
           WARNING  [17:50:47] Speaker 0 is not found. Use speaker 0 core.py:206
                    instead.                                                    
[17:50:48] WARNING  [17:50:48]                                   warnings.py:109
                    /usr/local/Caskroom/miniforge/base/lib/pytho                
                    n3.10/site-packages/torch/nn/functional.py:4                
                    697: UserWarning: The operator                              
                    'aten::im2col' is not currently supported on                
                    the MPS backend and will fall back to run on                
                    the CPU. This may have performance                          
                    implications. (Triggered internally at                      
                    /Users/runner/work/pytorch/pytorch/pytorch/a                
                    ten/src/ATen/mps/MPSFallback.mm:11.)                        
                      return torch._C._nn.im2col(input,                         
                    _pair(kernel_size), _pair(dilation),                        
                    _pair(padding), _pair(stride))                              
                                                                                
Traceback (most recent call last):
  File "/usr/local/Caskroom/miniforge/base/bin/svc", line 8, in <module>
    sys.exit(cli())
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/click/core.py", line 1130, in __call__
    return self.main(*args, **kwargs)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/click/core.py", line 1055, in main
    rv = self.invoke(ctx)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/click/core.py", line 1657, in invoke
    return _process_result(sub_ctx.command.invoke(sub_ctx))
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/click/core.py", line 1404, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/click/core.py", line 760, in invoke
    return __callback(*args, **kwargs)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/__main__.py", line 277, in infer
    infer(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/inference/main.py", line 95, in infer
    audio = svc_model.infer_silence(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/inference/core.py", line 296, in infer_silence
    audio_chunk_pad_infer_tensor, _ = self.infer(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/inference/core.py", line 221, in infer
    c, f0, uv = self.get_unit_f0(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/inference/core.py", line 149, in get_unit_f0
    f0 = so_vits_svc_fork.f0.compute_f0(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/f0.py", line 202, in compute_f0
    f0 = compute_f0_crepe(wav_numpy, p_len, sampling_rate, hop_length, **kwargs)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/f0.py", line 170, in compute_f0_crepe
    pitch: Tensor = torchcrepe.predict(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torchcrepe/core.py", line 120, in predict
    probabilities = infer(frames, model)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torchcrepe/core.py", line 558, in infer
    torchcrepe.load.model(frames.device, model)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torchcrepe/load.py", line 30, in model
    torch.load(file, map_location=device))
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 809, in load
    return _load(opened_zipfile, map_location, pickle_module, **pickle_load_args)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 1172, in _load
    result = unpickler.load()
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 1142, in persistent_load
    typed_storage = load_tensor(dtype, nbytes, key, _maybe_decode_ascii(location))
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 1116, in load_tensor
    wrap_storage=restore_location(storage, location),
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 1086, in restore_location
    return default_restore_location(storage, str(map_location))
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 220, in default_restore_location
    raise RuntimeError("don't know how to restore data location of "
RuntimeError: don't know how to restore data location of torch.storage.UntypedStorage (tagged with mps:0)

[Grownz]

I'm sorry but I can't recreate that error locally. But i found that it coud be a pytorch-related problem of DirecX for machine learning (microsoft/DirectML#196).

Did you try an fresh reinstall with all the necessary dependencies?

[noisyneil]

Did you try an fresh reinstall with all the necessary dependencies?

Thankyou for taking the time to reply. 🙏🏼

I haven't yet, but I will. I assumed everything was installed correctly as the GUI version is flawless. Hmm... I have a suspicion I'm still not formulating the command correctly. I launch Terminal and then run this:

svc infer -m "/Users/noisyneil/Documents/- MY DOCUMENTS/MUSIC/AI VOICE MODELS/F - Lana Del Rey 100k/G_100000.pth" -c "/Users/noisyneil/Documents/- MY DOCUMENTS/MUSIC/AI VOICE MODELS/F - Lana Del Rey 100k/config.json" -o "/Users/noisyneil/Desktop/AI Vocal Temp.wav" -fm crepe -d cpu "/Users/noisyneil/Desktop/Audio.wav"

[Grownz]

I can give you a line that worked for me in a try to recreate your problem yesterday for reference. I included special characters as the brackets, spaces and the french é to exclude these sources of error.

svc infer -m .\logs\44k -c .\configs\44k\config.json -o .\Champs_Frosch.wav -fm crepe -d cpu 'D:\FTP\AI-Shit\Soundsachen\Joe Dassin - Les Champs-Elysées\1_Joe Dassin - Les Champs-Elysées (Audio) (128kbit_AAC)_(Vocals).wav'

The main differences i can spot (w/o being an expert terminal power user xD ) are semantically that i don't refer to exact model files but relative directories with .\ (picks the latest model automatically) and syntactically that i use backslashes as well as single apostrophs. Maybe you can try that as well.

[noisyneil]

Ok I moved a model to the desktop, simplified its folder name and the output path so that spaces or symbols should be ruled out:

svc infer -m "/Users/noisyneil/Desktop/ModelTest/G_100000.pth" -c "/Users/noisyneil/Desktop/ModelTest/config.json" -o "/Users/noisyneil/Desktop/Output.wav" -fm crepe -d cpu "/Users/noisyneil/Desktop/Audio.wav"

Same result unfortunately!

I should also mention I'm on mac, so I'm not sure if your use of \ would be applicable to me...?

My lack of CLI fluency is very frustrating, as I'm so close to being able to run male backing vocals through female models in batches in half the time compared to using the GUI.

Thanks again! ♥️

[noisyneil]

Thanks for clearing that up. 🙏🏼

So I need to put the config file inside config/44k, in the same directory as svc.

What about the .pth file? I don't see that in the default structure list you shared.

[Grownz]

The .pth contains the model and is the training output. Therefore, it's located in the \logs\44k\.

In my example (after 3k training epochs):

To work with the completed model, you only need the G_xxxxx.pth. And don't get confused, the config file here is just for training purposes, not infernce.

The default paths for model and config are shown in the infer help, as well:

[noisyneil]

Ah man, I give up. 😩

I created the directories as you said:

/Users/noisyneil/Desktop/ModelTest/logs/44k/G_100000.pth
/Users/noisyneil/Desktop/ModelTest/configs/44k/config.json

Then I ran:

cd /Users/noisyneil/Desktop/ModelTest

...followed by:

svc infer -m ./logs/44k -c ./configs/44k/config.json -o ./Output.wav -fm crepe -d cpu "/Users/noisyneil/Desktop/Audio.wav"

...which, as far as I can see is identical in format to your previous example.

Here's the output in Terminal:

Click me

INFO     [13:14:24] Since model_path is a directory,  __main__.py:273
                    use logs/44k/G_100000.pth                                   
[13:14:25] WARNING  [13:14:25]                                   warnings.py:109
                    /usr/local/Caskroom/miniforge/base/lib/pytho                
                    n3.10/site-packages/so_vits_svc_fork/modules                
                    /synthesizers.py:81: UserWarning: Unused                    
                    arguments: {'n_layers_q': 3,                                
                    'use_spectral_norm': False}                                 
                      warnings.warn(f"Unused arguments:                         
                    {kwargs}")                                                  
                                                                                
           INFO     [13:14:25] Decoder type: hifi-gan        synthesizers.py:100
[13:14:27] INFO     [13:14:27] Loaded checkpoint                    utils.py:261
                    'logs/44k/G_100000.pth' (epoch 3508)                        
           INFO     [13:14:27] Chunk: Chunk(Speech: False, 8820.0)   core.py:283
           INFO     [13:14:27] Chunk: Chunk(Speech: True, 264600.0)  core.py:283
           WARNING  [13:14:27] Speaker None is not found. Use        core.py:206
                    speaker 0 instead.                                          
[13:14:28] WARNING  [13:14:28]                                   warnings.py:109
                    /usr/local/Caskroom/miniforge/base/lib/pytho                
                    n3.10/site-packages/torch/nn/functional.py:4                
                    697: UserWarning: The operator                              
                    'aten::im2col' is not currently supported on                
                    the MPS backend and will fall back to run on                
                    the CPU. This may have performance                          
                    implications. (Triggered internally at                      
                    /Users/runner/work/pytorch/pytorch/pytorch/a                
                    ten/src/ATen/mps/MPSFallback.mm:11.)                        
                      return torch._C._nn.im2col(input,                         
                    _pair(kernel_size), _pair(dilation),                        
                    _pair(padding), _pair(stride))                              
                                                                                
Traceback (most recent call last):
  File "/usr/local/Caskroom/miniforge/base/bin/svc", line 8, in <module>
    sys.exit(cli())
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/click/core.py", line 1130, in __call__
    return self.main(*args, **kwargs)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/click/core.py", line 1055, in main
    rv = self.invoke(ctx)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/click/core.py", line 1657, in invoke
    return _process_result(sub_ctx.command.invoke(sub_ctx))
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/click/core.py", line 1404, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/click/core.py", line 760, in invoke
    return __callback(*args, **kwargs)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/__main__.py", line 277, in infer
    infer(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/inference/main.py", line 95, in infer
    audio = svc_model.infer_silence(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/inference/core.py", line 296, in infer_silence
    audio_chunk_pad_infer_tensor, _ = self.infer(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/inference/core.py", line 221, in infer
    c, f0, uv = self.get_unit_f0(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/inference/core.py", line 149, in get_unit_f0
    f0 = so_vits_svc_fork.f0.compute_f0(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/f0.py", line 202, in compute_f0
    f0 = compute_f0_crepe(wav_numpy, p_len, sampling_rate, hop_length, **kwargs)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/so_vits_svc_fork/f0.py", line 170, in compute_f0_crepe
    pitch: Tensor = torchcrepe.predict(
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torchcrepe/core.py", line 120, in predict
    probabilities = infer(frames, model)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torchcrepe/core.py", line 558, in infer
    torchcrepe.load.model(frames.device, model)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torchcrepe/load.py", line 30, in model
    torch.load(file, map_location=device))
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 809, in load
    return _load(opened_zipfile, map_location, pickle_module, **pickle_load_args)
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 1172, in _load
    result = unpickler.load()
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 1142, in persistent_load
    typed_storage = load_tensor(dtype, nbytes, key, _maybe_decode_ascii(location))
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 1116, in load_tensor
    wrap_storage=restore_location(storage, location),
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 1086, in restore_location
    return default_restore_location(storage, str(map_location))
  File "/usr/local/Caskroom/miniforge/base/lib/python3.10/site-packages/torch/serialization.py", line 220, in default_restore_location
    raise RuntimeError("don't know how to restore data location of "
RuntimeError: don't know how to restore data location of torch.storage.UntypedStorage (tagged with mps:0)

You've been very generous with your help, so thankyou!

[noisyneil]

Hi. As it turns out, for some reason crepe doesn't work on my system. Not sure why.

A Keyboard Maestro forum user suggested I switch to dio, and the inference ran successfully. However, the pitch interpretation is horrible. I'm not sure why this is, as it's perfect via the GUI.

[Grownz]

ok, this is really weird. I'm glad tho, you found the issue 😀