Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Wav2vec-U Gan train WER UER loss are not satisfactory wav2vec-u wav2vec-u 2.0 gan #5572

Open
XR1988 opened this issue Dec 11, 2024 · 0 comments

Comments

@XR1988
Copy link

XR1988 commented Dec 11, 2024

I’m also trying to reproduce the results, but the WER and UER are not satisfactory (I’ve been stuck for two weeks now) and would like to ask for your guidance.

Modifications: The setup, environment, and models were built according to the original instructions, with no significant changes. Everything runs smoothly, and there are almost no errors.

Data:

LibriSpeech-100h
Librispeech-lm-norm.txt.gz
G2P phonemes were used for processing, and no errors occurred during preprocessing.
Current Status:
For both Wav2vec-U 1.0 and 2.0, the metrics (e.g., WER and UER) drop slightly in the first few minutes of training but then stagnate.

Attempts:

Tried different hyperparameter configurations.
Tested training on CPU or GPU.
Used larger datasets.
Extended training durations.
Unfortunately, the performance remains stagnant regardless of these efforts.

1.0
b74726492e12e6a4cc85aadb494e1e0

2.0
caf4f7b465790134dfc9e724912b6a1

data
8259d25567581ac80ecc012c81c3283

test resilt
image

Let me know if additional details are needed!
Looking forward to your advice.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

1 participant