linalg.eigh Convergence Issue in Speaker Diarization about whisper-diarization

[MahmoudAshraf97]

Hi, can you try to use the notebook provided and see what line causes that error?

[lamiranda-br]

Sure, error started on line 123 ˜msdd_model.diarize()˜.

I believe it was some dependency problem caused by installing all the requirements at once.
When I install line by line following the proposed sequence on requirements.txt, it works without errors.

Thank you!

Traceback (most recent call last) ──────────────────────╮
│ /content/drive/MyDrive/Colab Notebooks/whisper-diarization/diarize.py:123 in │
│
│ 121 # Initialize NeMo MSDD diarization model
│ 122 msdd_model = NeuralDiarizer(cfg=create_config())
│ ❱ 123 msdd_model.diarize()
│ 124
│ 125 del msdd_model
│ 126 torch.cuda.empty_cache()
│
│ /usr/local/lib/python3.9/dist-packages/nemo/collections/asr/models/msdd_mode
│ ls.py:1170 in diarize
│
│ 1167 │ │ function that can generate overlapping timestamps. `self.run_
│ 1168 │ │ """
│ 1169 │ │ torch.set_grad_enabled(False)
│ ❱ 1170 │ │ self.clustering_embedding.prepare_cluster_embs_infer()
│ 1171 │ │ self.msdd_model.pairwise_infer = True
│ 1172 │ │ self.get_emb_clus_infer(self.clustering_embedding)
│ 1173 │ │ preds_list, targets_list, signal_lengths_list = self.run_pair

[...]

│ /usr/local/lib/python3.9/dist-packages/nemo/collections/asr/parts/utils/offl │
│ ine_clustering.py:546 in eigValueSh │
│ │
│ 543 │ │ laplacian = laplacian.float().to(device) │
│ 544 │ else: │
│ 545 │ │ laplacian = laplacian.float() │
│ ❱ 546 │ lambdas = eigvalsh(laplacian) │
│ 547 │ return lambdas │
│ 548 │
│ 549 │
╰──────────────────────────────────────────────────────────────────────────────╯
_LinAlgError: linalg.eigh: The algorithm failed to converge because the input
matrix is ill-conditioned or has too many repeated eigenvalues (error code:
842).