List your company

Audio Watermarking Robustness on LibriSpeech and Common Voice (test)

100No Attack RobustnessAudioSeal
83.3687.689296.32Nov 4, 2025

Evaluation results

MethodLinks
AudioSealmode=Single Watermark, abbreviation=A2025.11100857085010010015940982251009864.8
SilentCiphermode=Single Watermark, abbreviation=S2025.11100751321174577731527216621912345.7
Para. A+Pstrategy=Parallel Addition, components=AudioSeal + PerTh2025.111009179848110010079970865372989280.8
Seq. A-Pstrategy=Sequential Cascading, order=AudioSeal then PerTh2025.1110089708577929074820845935939175.3
Seq. P-Astrategy=Sequential Cascading, order=PerTh then AudioSeal2025.1110090758178100100719709867691008770.7
Para. A+Sstrategy=Parallel Addition, components=AudioSeal + SilentCipher2025.11100897393201001007891259474391005474.7
Para. P+Sstrategy=Parallel Addition, components=PerTh + SilentCipher2025.1110085423178981001004226319364957680.9
PA-TFMstrategy=Unified Multiplexing2025.1110091818582981001006216719363969882.4
MaskNetstrategy=Unified Multiplexing2025.111009585808998100100961892904110010085.6
PerThmode=Single Watermark, abbreviation=P2025.118466378768478762014735506748