AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

Detect All-Type Deepfake Audio: Wavelet Prompt Tuning for Enhanced Auditory Perception

Yuankun Xie, Ruibo Fu, Xiaopeng Wang, Zhiyong Wang, Songjun Cao, Long Ma, Haonan Cheng, Long Ye

AAAI 2026 · Vision-language and multimodal › Cross-modal learning · approach: Audio-visual speech separation

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add wavelet prompt tokens (4 WPT + 6 PT) to prompt tuning (ablation)PT-XLSR-AASIST · speech-trained, avg EER over 4 typesEER (%): 30.79 -> 26.86Improved
use WPT instead of full fine-tuning with 458x fewer trainable parametersXLSR-AASIST · speech-trainedEER (%): 27.68 -> 26.86Improved
co-train on all four audio types with WPTXLSR-AASIST · all-type ADDavg EER (%): 4.98 (FT) -> 3.58 (WPT)Improved
apply WPT to other SSL front-endsWavLM-AASIST, MERT-AASIST · all-type ADD co-trainedavg EER (%): 10.81 -> 7.66 (WavLM); 13.21 -> 9.70 (MERT)Improved
increase prompt tokens from 10 to 100 or 200 (ablation)PT-XLSR-AASIST · speech-trainedavg EER (%): 30.79 -> 31.28 (100), 33.36 (200)Hurt
insert prompts in every layer instead of first layer only (ablation)PT-XLSR-AASIST · speech-trainedavg EER (%): 32.54 -> 30.79Improved
keep prompt token in last layer instead of deleting it (ablation)PT-XLSR-AASIST · speech-trainedavg EER (%): 33.07 -> 30.79Improved
use all 10 tokens as wavelet tokens (ablation)WPT-XLSR-AASIST · speech-trainedavg EER (%): 26.86 -> 32.54Hurt
fine-tune SSL instead of freezing it on music-trained CMs (ablation)XLSR/WavLM-AASIST · music-trainedEER (%): FT-XLSR avg 45.91 vs FR-XLSR avg 35.63Hurt
fine-tune SSL instead of freezing it on speech, sound, singing-trained CMsXLSR-AASIST · in-domain EEREER (%): 1.28 -> 0.38 (speech); 2.40 -> 0.21 (sound)Improved
train on a single audio typeall CMs · cross-type ADDEER (%): speech-trained AASIST: 1.48 in-domain vs 48.32 on soundHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy