Compairing RTTM file

[aishwaryjoshi31]

Hi,

I have successfully diarized a meeting from AMI corpus. I want to check the diarization error that my output contains. How do I check? I checked AMI corpus but did not find rttm file there to compare with rttm file that is generated as a result of diarization. Can anybody help?

[mrsrikanth]

Hello,

You will have to create your own RTTM files from groundtruth
available in the AMI dataset. A sample RTTM file is already available
in the package.

Srikanth

[aishwaryjoshi31]

Hi @mrsrikanth Can you please elaborate more. I did not get you.

[mrsrikanth]

The AMI dataset has manual annotations that has
segment-level information (in XML format) for each speaker for each recording.
You can easily convert this to RTTM format to use as groundtruth for
diarization.

Srikanth

[aishwaryjoshi31]

The segment folder contains information about each recording only , I did not see speaker information in that folder so here is image of one of the xml file in segment folder. Can you point it out where is speaker information?

[mrsrikanth]

The file name that you have attached is IS1000a.A.segments.xml.
The 'A' in the suffix is the speaker name for that file. You will also find
segment information for speakers B, C and D. To translate A, B, C and D
to global-level speaker identities, you can check .corpusResources/meetings.xml.

[aishwaryjoshi31]

Thanks @mrsrikanth for your help. Few last things : Are there any overlapping between the speakers, for eg speaker s1 is speaking from let say t=1 to t=5 and speaker s2 speaking from t=3 to t=7. So are there any such cases like this one?

[cunnie]

@aishwaryjoshi31 You may find a script I wrote useful for converting the AMI Corpus's XML files to RTTM format. Typical invocation:

nite_xml_to_rttm.py ~/Downloads/ami_public_manual_1.6.2/words/ES2008a.*.words.xml |
    sort -n -k 4 > /tmp/ES2008a.rttm