About maximum duration of segments

[enisimsar]

Hi all,

Thanks for the great library.

I am trying to segment audio with the dia pipeline.

import torch
from pyannote.audio.features import RawAudio

pipeline = torch.hub.load("pyannote/pyannote-audio", "dia")

OWN_FILE = {"audio": input_file}

diarization = pipeline(OWN_FILE)
segments = diarization.for_json()["content"]

However, I need to specify minimum and maximum durations for segments. I checked the parameters of the parts of the pipeline and could not find a solution. Do you have any idea?

[hbredin]

There is no easy way to do that (though I am not sure why you would need to specify a maximum duration).

FYI, pyannote.audio 1.x will soon be superseded by pyannote.audio 2.0 with a much better speaker diarization pipeline, which has a min_duration_on hyper-parameter that could you use to set the minimum duration of speech turns.

[enisimsar]

Thanks a lot, it worked. I have tried some options for min_duration_on. Now, I can divide the longer segments.

I will also change my code to 2.0.

[Yagna24]

Hi @enisimsar , thank you for asking this question, this also helps me.
I have a doubt, where does the input audio .wav file processes first i.e. what .py file is responsible to fetch the audio file when we use

OWN_FILE = {"audio": input_file}

diarization = pipeline(OWN_FILE)