AMI diarization

I’m trying to follow the recipe for speaker diarization on the AMI dataset (https://github.com/speechbrain/speechbrain/tree/develop/recipes/AMI/Diarization) but unfortunately without success. Here’s the output:

...
speechbrain.utils.parameter_transfer - Loading pretrained files for: embedding_model, mean_var_norm_emb
__main__ - Tuning for p-value for SC (Multiple iterations over AMI Dev set)
__main__ - Diarizing dev set
__main__ - No recording IDs found! Please check if meta_data json file is properly generated.

I have downloaded the data and set the variables in the config files accordingly, i.e.:

data_folder: .../AMI/amicorpus/
manual_annot_folder: .../AMI/ami_public_manual_1.6.2 

where amicorpus looks as follows: amicorpus/EN2009d/audio/EN2009d.Mix-Headset.wav

I’m running this using device: 'cpu'

I checked the results/…/metadata folder and I see that ami_dev.Mix-Headset.subsegs.json and eval.Mix-Headset.subsegs.json are empty, while ami_train.Mix-Headset.subsegs.json contains a dict of elemts like

"EN2009d_0.0_2.99": {
    "wav": {
      "file": "/Users/jonas/Desktop/Translated/ASR/datasets/AMI/amicorpus//EN2009d/audio/EN2009d.Mix-Headset.wav",
      "duration": 2.99,
      "start": 0,
      "stop": 47840
    }
  }

I would really appreciate some help! Am I missing anything?

[nauman-daw]

Hi,

Yes. “No recording IDs found! Please check if meta_data json file is properly generated.” should be related to improper paths. Please check “<filename>.subsegs.json” as this will be by your experiment.py.

@LONG520520 please feel free to open a PR for this, I will check it. Even if the PR suggests some useful points on how to avoid these path errors, it will be very helpful for others.

thank you very much!

[nauman-daw]

Closing this path issue for now.