preprocessing: only decode encoded filenames

NVIDIA / OpenSeq2Seq

Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP

https://nvidia.github.io/OpenSeq2Seq

Apache License 2.0

1.54k stars 369 forks source link

preprocessing: only decode encoded filenames #314

Closed mvankeirsbilck closed 5 years ago

mvankeirsbilck commented 5 years ago

In some cases (I'm not sure why) filenames are read as binary strings (python 'bytes' type) from the csv. This required conversion to normal strings using the .decode() function. If the filename is already a normal string (as should be the default case), this conversion is not required and in fact throws an error if attempted.