A Novel Dataset for Testing Anti-spoofing Models in a Telephony Environment.

ACL ARR 2024 December Submission742 Authors

Abstract: In the last few years, synthetic voices have become incredibly realistic and more difficult to discriminate from authentic, human voices. Although impressive, these advances raise concerns about safety and security, increasing the need for models that can discriminate between human and synthetic voices under realistic conditions. While previous work has created datasets and models that provide convincing results for high quality recordings, it is unclear how well they generalize to different conditions. In this paper, we present a novel dataset for testing the performance of anti-spoofing models in noisy conditions associated with the cellular telephone network. We demonstrate that a model trained on this dataset can achieve high accuracy on this novel telephony data without any degradation in accuracy on non-telephonic audio.

Paper Type: Short

Research Area: Speech Recognition, Text-to-Speech and Spoken Language Understanding

Research Area Keywords: automatic speech recognition; speech technologies

Contribution Types: Model analysis & interpretability, Publicly available software and/or pre-trained models, Data resources

Languages Studied: English

Submission Number: 742