search

eurecom-asp / rawnet2-antispoofing

This repository includes the code to reproduce our paper "End-to-end anti-spoofing with RawNet2" (https://arxiv.org/abs/2011.01108) published in ICASSP '21.
MIT License
49 stars 13 forks source link

A problem of input and output shape of the model architecture #5

Open Yang-son opened 1 year ago

Yang-son commented 1 year ago

Dear author:

I have been working on reproducing the results. While going through the code provided in the repository, I noticed that there is an inconsistency between the model architecture in the code and the description provided in the paper.

In particular, the input (number of samples) and output shape of the model in the code do not match with the ones mentioned in the paper. This is causing some confusion for me as I am trying to reproduce the results from the paper.

image 2023-04-15_095355

I appreciate the hard work put in by the author to make their code available and I am grateful for that.

Would it be possible for the author to clarify the inconsistency between the model structure in the code and the paper, and provide any guidance on how to resolve this issue?

Thank you for your attention to this matter.

TakHemlata commented 1 year ago

Hi Yang-son,

Thanks for your interest.

The shape is changed because of input samples. I actually updated the repo and copied the config file from the RawNet2 baseline for ASVspoof 2021 challenge in which we used input samples=64600. I again changed it to 64000 as mentioned in the paper. If you want you can also check the RawNet2 baseline code from (https://github.com/asvspoof-challenge/2021/tree/main/LA/Baseline-RawNet2).

Please let me know if you need any help.

Thanks

Yang-son commented 1 year ago

Thank you so much for your response to my previous issue.

Your help was greatly appreciated!

I have a follow-up question regarding the training data partitioning mentioned in the paper. Specifically, the paper states that "We re-partitioned the data by supplementing training data with 90% of development data. The remaining 10% was used for validation."

image

However, I could not find the code that generates the new training data by adding 90% of the development data.

Thank you in advance for your time and assistance!