Skip to content

Performance discrepancy on LRS3 with model vsr_trlrwlrs2lrs3vox2avsp_base.pth #57

Description

@aizhiqi-work

Description:
When using the model vsr_trlrwlrs2lrs3vox2avsp_base.pth, the performance on the LRS3 dataset is significantly lower than expected. After processing the data in the same way as before, the model's word error rate (WER) on the LRS3 dataset is around 23-24%. The final result indicates:
================ Final Result ================
Total samples found : 1321
Valid samples : 1321
Skipped samples : 0
Total words (N) : 9890
Substitutions (S) : 1667
Deletions (D) : 407
Insertions (I) : 214
WER : 0.231345

Steps to Reproduce:

Use the model vsr_trlrwlrs2lrs3vox2avsp_base.pth.
Process the same data in the same way as the previous steps.
Evaluate the model on the LRS3 dataset.

Has anyone encountered a similar situation?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions