Nemo-Auto-Subtitle

Using Nvidia NeMo to auto create subtitles, many thanks to @RaviSoji for the PLDA!

How it works

Basically, do VAD and find the timings, perform STT then NMT. However, the VAD from Nemo doesn’t seem to give satisfactory results. Instead a diarization model is used to extract the embeddings, which then calls for the use of PLDA to predict speech segments from those.

Need to put relevant data into plda_data/voiceand plda_data/background.

Note: Nemo doesn’t work well on Windows due to pyannote’s multiprocessing. Modify the _init_.py of pyannote.metrics (commented out this line). Seems to work without issue.

# manager_ = Manager()

Usage:

TODO, whole working thing is in speakernet_playground.py at the moment.