How to save and load diarization results ?

[amitli1]

I want to save the pyannote/speaker-diarization results to file (text or pickle) and load it later (instead of rerunning the diarization pipeline).

Is it possible ?

pipeline   = Pipeline.from_pretrained("pyannote/speaker-diarization")
diarization = pipeline("test.wav")

I want to save diarization

I saw that I can save it as rttm file, but don't have the opportunity to load this rttm to pyannote.core.annotation.Annotation

[diagomike]

this is how you write to rttm file

with open('out_put.rttm','w') as f:
  groundtruth.write_rttm(f)

[rmatinha-fctunl]

It says that the 'DiarizeOutput' object has no attribute 'write_rttm'

[hbredin]

diarization.speaker_diarization.write_rttm(...)

[rmatinha-fctunl]

Thank you so much! It worked! I had opened a discussion with a similar problem before, do you mind if I paste this solution there?

[diagomike]

this will save index [start -> end] format into a simple text, which you can easily write to disk

simple_text = ''
for (start,end),index in groundtruth.itertracks():
  simple_text += f"%s: [%.2fs -> %.2fs]\n" % (index,start, end)

[mcho421]

It looks like there's support to read and write .rttm or .lab files. Here's an example using .lab:

# Writing to a file on disk
diarization.write_lab('result.lab')

# Writing to a string
result = diarization.to_lab()
print(result)

# Loading the file
from pyannote.database.util import load_lab
diarization = load_lab("result.lab")