[Bug] Multispeaker VITS training does not work in direct python but does with config.json

Describe the bug

Hi,

I want to train VITS model with mutliple speakers and external embeddings (aka d_vectors). So I provided VitsArgs in a recipe :

vits_args = VitsArgs(
    use_language_embedding=False,
    embedded_language_dim=1,
    use_speaker_embedding=False,
    num_languages=1,
    use_sdp=False,
    #Those 3 properties also have to be repeated in the config section (see https://github.com/coqui-ai/TTS/issues/1454#issuecomment-1081843205)
    use_d_vector_file=True, 
    d_vector_file="/home/Caraduf/Models/d_vector_file_4_Voices.json",
    d_vector_dim=512
)

and repeated the same 3 lines in the VitsConfig section as explained here :

config = VitsConfig(
model_args = vits_args,
    use_d_vector_file=True, 
    d_vector_file="/home/Caraduf/Models/d_vector_file_4_Voices.json",
    d_vector_dim=512

I then ran the training python3 my_multispeaker_vits_training.py but it failed due to

 line 303, in _conv_forward
    return F.conv1d(input, weight, bias, self.stride,
 TypeError: conv1d() received an invalid combination of arguments - got (NoneType, Parameter,    Parameter, tuple, tuple, tuple, int), but expected one of:
  * (Tensor input, Tensor weight, Tensor bias, tuple of ints stride, tuple of ints padding, tuple of ints dilation, int groups)
      didn't match because some of the arguments have invalid types: (NoneType, Parameter, Parameter,  tuple, tuple, tuple, int)
  * (Tensor input, Tensor weight, Tensor bias, tuple of ints stride, str padding, tuple of ints dilation, int groups)
      didn't match because some of the arguments have invalid types: (NoneType, Parameter, Parameter, tuple, tuple, tuple, int)

This is also described by @harmlessman in this comment.

But if I run the train_tts.py --config_path path/to/the/jus/previously/created/config.json then the multispeaker training with external embeddings runs fine.

To Reproduce

See above.

Expected behavior

Launching a multi speaker training via “direct python way” providing the external speaker embeddings should work directly without the need to use the generated config.json file via generic python3 training_tts.py --config_path X/Y/Z/config.json.

Logs

No response

Environment

CoquiTTS 0.8.0

Additional context

No response

[shigabeev]

Hi! Did you find a solution for an issue? I’m facing the same one unfortunately.

[Ca-ressemble-a-du-fake]

That was what I was missing ! Thanks for sharing your solution @lokmantsui this is much better than my workaround !