Audio Splicing Localization: Can We Accurately Locate the Splicing Tampering?
Zhiping Zeng, Zhizheng Wu · 2022 13th International Symposium on Chinese Spoken Language Processing (ISCSLP) · 2022
Audio splicing is a low-cost and straightforward tampered form of audio fraud. Splicing localization is to locate where the splicing tampering happens. Although there are a few studies on audio splicing localization, most of them use a much smaller dataset and locate the splicing points within a long segment, which can not accurately reflect the real model performance in the real-world application. In this study, we revisit audio splicing localization by using a large-scale splicing dataset and how accurate the model can be if we narrow down the localization range. We employ a ResNet-based model for localization and develop a large-scale dataset from LibriSpeech for experiments. The dataset consists of more than 130k utterances, including insertion, deletion and substitution splicing tampering.