verifification.py

[Adarsh474532]

sir speaker is different but verify is correct, how sir.

[Adarsh474532]

result is wrong

[jymsuper]

The actual result is not wrong. It just shows wrong name. It says 'Claimed speaker : test'. Here, the 'test' is the name of the feature folder.

Are you using Window OS? Then, please change the code as follows (I assumed that the code is working on Ubuntu OS).

In the function 'perform_verification' of 'verification.py', Please change the code as below (use os.sep instead of '/'). test_spk = testfilename.split(os.sep)[-2].split('')[0]

[Adarsh474532]

This time is also wrong.

[Adarsh474532]

but test with another embedding then result is right, how

[jymsuper]

Did you change what I wrote above? In your screen, claimed speaker is 'feat' True speaker or Claimed speaker should be the name of the speaker folder. Please carefully check the function 'perform_verification' and debug yourself. I guess this is just the print issue.

[Adarsh474532]

def perform_verification(use_cuda, model, embeddings, enroll_speaker, test_filename, test_frames, thres): enroll_embedding = embeddings[enroll_speaker] test_embedding = get_embeddings(use_cuda, test_filename, model, test_frames)

score = F.cosine_similarity(test_embedding, enroll_embedding)
score = score.data.cpu().numpy() 

if score > thres:
    result = 'Accept'
else:
    result = 'Reject'

test_spk = test_filename.split(os.sep)[-2].split('_')[0]
print("\n=== Speaker verification ===")
print("True speaker: %s\nClaimed speaker : %s\n\nResult : %s\n" %(enroll_speaker, test_spk, result))
print("Score : %0.4f\nThreshold : %0.2f\n" %(score, thres))

[Adarsh474532]

but , with old speaker work is correctly.

[jymsuper]

The folder structure should be the same. Check the folder structure. Also, check the variable 'test_spk' and 'test_filename'.

[Adarsh474532]

my structure

[Adarsh474532]

speaker verification.docx

Check the code please

[jymsuper]

I don't know why 'feat' is printed for 'Claimed speaker'. Check the variable 'test_spk' and 'test_filename'.

[Adarsh474532]

speaker verification.docx

Check the code please

code is correct or wrong

[jymsuper]

I think the instruction is enough. Please debug yourself. This is just basic python debugging, not related to speaker verification algorithm. As I already mentioned, I guess this is just the print issue. Please check the variable 'test_spk' and 'test_filename'.

[Adarsh474532]

error in sr_dataset.py

[Adarsh474532]

I think the instruction is enough. Please debug yourself. This is just basic python debugging, not related to speaker verification algorithm. As I already mentioned, I guess this is just the print issue. Please check the variable 'test_spk' and 'test_filename'.

ok sir

[jymsuper]

You can ignore the main function in 'SR_dataset.py'. The main function is not used at all. Another functions in 'SR_dataset.py' are used in other files such as 'train.py'

[Adarsh474532]

ok

[Adarsh474532]

one last question sir

np.set_printoptions(threshold=np.nan)

this line

[jymsuper]

You can remove the line

[Adarsh474532]

ok

[Adarsh474532]

Thank you sir for helping, project is not complete , my hope is down.

Thank you for helping

[Adarsh474532]

sir, in the ubuntu same problem result is wrong

[Adarsh474532]

sir, it is possible any error in feature extraction then files is create error.

[jymsuper]

There should be no errors in feature extraction