luferrer / dca-plda Goto Github PK

Hi, I want to train and evaluate the model on custom data. Can you explain the structure of the eval files from the example data, I am not 100% sure what needs to be the content of .lst files. enroll.lst and test.lst are identical in the provided example.

The format is not hdf5 or npz using kaldi so I am not sure what to change in the code.

Thank you very much!

Hi,
Thank you for your paper and repos. I think this is an astonish innovation for speech verification. I am an undergraduate student trying to train the model with our mother langue dataset. But we have an error and cannot figure it out why. Could you guys have look at the errors and have some suggestion how to fix it for us?

Traceback (most recent call last):
File "expe.py", line 196, in
trn_loss = model.init_params_with_data(trn_dataset, config_trn, device=device, subset=init_subset)
File "/home/thule/Project/voice-verification/DCA-PLDA-master/examples/modules.py", line 183, in init_params_with_data
self.lda_stage.init_with_lda(x, speaker_ids, init_params, sec_ids=domain_ids)
File "/home/thule/Project/voice-verification/DCA-PLDA-master/examples/modules.py", line 598, in init_with_lda
evals, evecs = linalg.eigh(BCov, WCov)
File "/home/thule/anaconda3/envs/Hope/lib/python3.8/site-packages/scipy/linalg/decomp.py", line 581, in eigh
raise LinAlgError('The leading minor of order {} of B is not '
numpy.linalg.LinAlgError: The leading minor of order 20 of B is not positive definite. The factorization of B could not be completed and no eigenvalues or eigenvectors were computed.

Our info in eigh function was 532 when it's supposed to be 0.
For x vector, we tried both MFCC and PNCC, but we had the same errors. Our xvector was ways smaller than yours though (about a hundredth of yours).

We are really lost. We really appreciate your helps.
Thank you for your time.

Hi,

I would like to train the plda with embeddings from my speaker verification model.
Can you give some insights on how to create train, val and eval data.

Thank You

Hi ,
i was following the steps you mentioned in README.md file and was trying to just run the run_all script but this following issue is coming up to due to missing file format_ld_results

Loaded key file data/eval/crowdsource/test.4_8_16.key
Loaded key file data/eval/crowdsource/test.32.key
./run_all: line 62: ../scripts/format_ld_results: No such file or directory

Thanking you in anticipation of your positive response,
Somil Jain

When I try to download the example data, after it is complete, the data is saved as .html webpage instead of tar.gz file, why?

I am having some issues downloading the sample data you provided. It seems the download always gets stuck after 460 MB for me. Not sure what's the issue, I have a stable internet connection.

Try this:

scores = np.array([1.0,2.0,3.0])
labels = np.array([0,0,1])
rocch = ROCCH(PAV(scores,labels))
ber, pmiss, pfa = rocch.Bayes_error_rate(-np.inf,True)

This gives ber, pmiss, pfa = (0,0,0), but pfa should be 1. The ROCCH is correctly computed and so is ber. But sometimes pmiss or pfa is not. The problem is here. The relevant vertex of the ROCCH is found by minimization. But sometimes, the minimum is not unique. The value of the minimum is the Bayes error-rate ber, which is then correct. But the wrong index sometimes happens to be chosen, returning the wrong pmiss or pfa.

I will figure out how to fix this.

Niko