running deeppavlov transcribe

Hi everybody,

I am trying to run a transcribe model on google colab, as:

from deeppavlov import build_model, configs

model = build_model(configs.nemo.asr)
text_batch = model("my_path")

I had an initial problem with model=build_model(configs.nemo.asr) because model nemo was not found, even if it was installed. Thus I solved this issue installing nemo from Nvidia repository as: !pip install git+https://github.com/NVIDIA/NeMo.git However, I am facing this new issue:

2020-05-11 10:47:37.536 ERROR in 'deeppavlov.core.common.params'['params'] at line 112: Exception in <class 'deeppavlov.models.nemo.asr.NeMoASR'>
Traceback (most recent call last):
  File "/usr/local/lib/python3.6/dist-packages/deeppavlov/core/common/params.py", line 106, in from_params
    component = obj(**dict(config_params, **kwargs))
  File "/usr/local/lib/python3.6/dist-packages/deeppavlov/models/nemo/asr.py", line 163, in __init__
    **self.nemo_params['AudioToMelSpectrogramPreprocessor']
  File "/usr/local/lib/python3.6/dist-packages/nemo/collections/asr/audio_preprocessing.py", line 362, in __init__
    super().__init__(n_window_size, n_window_stride)
  File "/usr/local/lib/python3.6/dist-packages/nemo/collections/asr/audio_preprocessing.py", line 74, in __init__
    super().__init__()
  File "/usr/local/lib/python3.6/dist-packages/nemo/backends/pytorch/nm.py", line 136, in __init__
    NeuralModule.__init__(self, name)  # For NeuralModule API
  File "/usr/local/lib/python3.6/dist-packages/nemo/core/neural_modules.py", line 70, in __init__
    super().__init__()
  File "/usr/local/lib/python3.6/dist-packages/nemo/core/neural_interface.py", line 41, in __init__
    self._app_state = nemo.utils.app_state.AppState()
  File "/usr/local/lib/python3.6/dist-packages/nemo/utils/metaclasses.py", line 36, in __call__
    cls.__instances[cls] = super(Singleton, cls).__call__(*args, **kwargs)
  File "/usr/local/lib/python3.6/dist-packages/nemo/utils/app_state.py", line 43, in __init__
    self._device = nemo.core.DeviceType.GPU
AttributeError: module 'nemo' has no attribute 'core'

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-54-5e369d1d15ef> in <module>()
----> 1 model = build_model(configs.nemo.asr)

9 frames
/usr/local/lib/python3.6/dist-packages/nemo/utils/app_state.py in __init__(self, device)
     41         # Had to set it to None in argument to avoid circular import at the class initialization phase.
     42         if device is None:
---> 43             self._device = nemo.core.DeviceType.GPU
     44         else:
     45             self._device = device

AttributeError: module 'nemo' has no attribute 'core'

Should I install nemo in some other way or is there a solution for this? Thanks very much

Stefano

[Steboss]

Ouch that’s great. Now it works fine I just changed the title of the issue, so it’s better and more related to what discussed. Thanks very much @IgnatovFedor I am closing the issue

[IgnatovFedor]

@Steboss, as mentioned in documentation, models work with single channel wav files. This error is most likely occurs due to your file is multi-channel. If linee_test.wav is multi channel, you could create single channel version using following code:

from scipy.io.wavfile import read, write
sr, data = read('linee_test.wav')
with open('linee_test_mono.wav', 'wb') as fout:
    write(fout, sr, data[:, 0])