Speech is Silver, Silence is Golden: What do ASVspoof-trained Models Really Learn?
share
We present our analysis of a significant data artifact in the official 2019/2021 ASVspoof Challenge Dataset. We identify an uneven distribution of silence duration in the training and test splits, which tends to correlate with the target prediction label. Bonafide instances tend to have significantly longer leading and trailing silences than spoofed instances. In this paper, we explore this phenomenon and its impact in depth. We compare several types of models trained on a) only the duration of the leading silence and b) only on the duration of leading and trailing silence. Results show that models trained on only the duration of the leading silence perform particularly well, and achieve up to 85 between 0 and 1). At the same time, we observe that trimming silence during pre-processing and then training established antispoofing models using signal-based features leads to comparatively worse performance. In that case, EER increases from 0.03 (with silence) to 0.15 (trimmed silence). Our findings suggest that previous work may, in part, have inadvertently learned thespoof/bonafide distinction by relying on the duration of silence as it appears in the official challenge dataset. We discuss the potential consequences that this has for interpreting system scores in the challenge and discuss how the ASV community may further consider this issue.
READ FULL TEXT