Audio segmentation of radio news and talk show in Non-English language.

[csikasote]

Hello, I have a set of audio recordings of radio news and talk show in an African language that I want to automatically segment. The audio durations range from 15 minutes to about 1hr each. Can I use pyannote to automatically segment them? I would to use the segments for unsupervised speech recognition experiments.

Any help will be appreciated.

[hbredin]

What did you try?

I'd recommend you start by using pyannote/segmentation for voice activity detection.

[csikasote]

Thank you for the recommendation. I was able to run the segmentation for VAD successfully. I have, however, two questions:

1. How do I generate 'segmentation' file that contains all the start and end of each segment.
2. How do I go about making sure that the segments are not more than 20 seconds?

I`m using pyannote for the first time.

[hbredin]

1. pyannote.audio pipelines return pyannote.core.Annotation instances that can be iterated over to get the segment start time and end time. See TL;DR section.
2. You would have to postprocess the segments yourself.