KeyError : No AudioToMelSpectrogramPreprocessor in 2_Online_ASR_Microphone_Demo.ipynb

[ierezell]

• Wanted behaviour :
  Run the notebook following given instructions

• Actual Behaviour :
  In model_definition['AudioToMelSpectrogramPreprocessor']['dither'] = 0
KeyError: 'AudioToMelSpectrogramPreprocessor'

There is no field AudioToMelSpectrogramPreprocessor in the yaml, thus all the subfields leads to error (same for model_definition['AudioToMelSpectrogramPreprocessor']['features'],
model_definition['AudioToMelSpectrogramPreprocessor']['pad_to'])

• Checkpoints taken from the comment in the notebook :
  the checkpoints are available from NGC: https://ngc.nvidia.com/catalog/models/nvidia:quartznet15x5

(tried with other files as well with no success)

[redoctopus]

Looks like the checkpoint has an older version of the quartznet15x5 config, and the class name has changed. Either one of the following should fix the error:

1. Use the config at examples/asr/configs/quartznet15x5.yaml, or
2. Change the key AudioPreprocessing to AudioToMelSpectrogramPreprocessor in the YAML file you are currently using.

[ierezell]

Thanks for the quick response !

Fixed by taking the config from github, the notebook is working (still have a pulseaudio bug on my side).