cl-fad-ver0

This repo is used to host the configs and checkpoints used in our ICASSP2024 submission projects.

How to run

The codebase is work-in-progress.

Exp Description

For each model, there are four exp setups, as shown in the following table.

For example, exp1.1 refers to an experiment where:

• The genuine data comes from Librispeech.
• The training order is organized by attacker's training dataset's name.
• The training strategy is finetuning.

<table> <tr> <th rowspan="1">Genuine Data</th> <th colspan="2">Librispeech</th> <th colspan="2">MLAAD</th> </tr> <tr> <th>Task Order</th> <th>By Dataset</th> <th>Random</th> <th>Random</th> </tr> <tr> <td>Finetune</td> <td>exp1.1</td> <td>exp1.2</td> <td>exp1.4</td> </tr> <tr> <td>One-Class Softmax</td> <td>exp2.1</td> <td>exp2.2</td> <td>exp2.4</td> </tr> <tr> <td>Replay</td> <td>exp3.1</td> <td>exp3.2</td> <td>exp3.4</td> </tr> <tr> <td>LwF</td> <td>exp4.1</td> <td>exp4.2</td> <td>exp4.4</td> </tr> </table>