verification PLDA

[zang5xixi]

when I finish the trianing of x-vector by 'python train_speaker_embeddings.py hparams/train_x_vector.yaml', i want to do the verification task by PLDA:'python speaker_verification_plda.py hparams/verification_plda_xvector.py', but i got a problem：no such file : mean_var_norm_emb.ckpt.

[mravanelli]

Yes, this is a known problem that we will fix as soon as we can. For now, You can save the mean_var_norm_emb vector at the end of train_speaker_embeddings.py (using torch.save)

On Sun, 26 Sept 2021 at 04:58, zang5xixi @.***> wrote:

when I finish the trianing of x-vector by 'python train_speaker_embeddings.py hparams/train_x_vector.yaml', i want to do the verification task by PLDA:'python speaker_verification_plda.py hparams/verification_plda_xvector.py', but i got a problem：no such file : mean_var_norm_emb.ckpt.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/speechbrain/speechbrain/issues/1011, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEA2ZVRQMHXMSM5SMKBGSJDUD3OCDANCNFSM5EYRVWCQ .

[zang5xixi]

Yes, this is a known problem that we will fix as soon as we can. For now, You can save the mean_var_norm_emb vector at the end of train_speaker_embeddings.py (using torch.save) … On Sun, 26 Sept 2021 at 04:58, zang5xixi @.***> wrote: when I finish the trianing of x-vector by 'python train_speaker_embeddings.py hparams/train_x_vector.yaml', i want to do the verification task by PLDA:'python speaker_verification_plda.py hparams/verification_plda_xvector.py', but i got a problem：no such file : mean_var_norm_emb.ckpt. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1011>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEA2ZVRQMHXMSM5SMKBGSJDUD3OCDANCNFSM5EYRVWCQ .

sorry, i am new to speaker recognition and it is my first time to use speechbrain. i do not know how to fix it.

[Drawlone]

It looks like that trainging process didn't use mean_var_norm_emb after computing embeeding. So it is diffcult for me to save it using torch.save. https://github.com/speechbrain/speechbrain/blob/1d194bfc51ae20b9e38596d220cdf0f4977e69de/recipes/VoxCeleb/SpeakerRec/train_speaker_embeddings.py#L72-L75 https://github.com/speechbrain/speechbrain/issues/1011#issuecomment-927267928

[mravanelli]

Sorry for the wrong suggestion. I think what you can do for now is to just remove the mean_var_norm_emb entry in the pretrainer object defined in the YAML file ( https://github.com/speechbrain/speechbrain/blob/develop/recipes/VoxCeleb/SpeakerRec/hparams/verification_plda_xvector.yaml#L81 and https://github.com/speechbrain/speechbrain/blob/develop/recipes/VoxCeleb/SpeakerRec/hparams/verification_plda_xvector.yaml#L84). This way the normalizer will be computed on-the-fly during the speaker_verification_plda.py. Please, let me know if it works that way.

On Mon, 27 Sept 2021 at 09:55, Drawlone @.***> wrote:

It looks like that trainging process didn't use mean_var_norm_emb after computing embeeding. So it is diffcult for me to save it using torch.save.

https://github.com/speechbrain/speechbrain/blob/1d194bfc51ae20b9e38596d220cdf0f4977e69de/recipes/VoxCeleb/SpeakerRec/train_speaker_embeddings.py#L72-L75

1011 (comment)

https://github.com/speechbrain/speechbrain/issues/1011#issuecomment-927267928

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/speechbrain/speechbrain/issues/1011#issuecomment-927898477, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEA2ZVVLKZNTZRPA4HQ4HWTUEBZUFANCNFSM5EYRVWCQ .

[Adel-Moumen]

Hello,

Any news regarding this issue please?

Thanks. Adel