DeepSonar: Towards Effective and Robust Detection of AI-Synthesized Fake Voices

View on arXiv ← Back to list

Authors: Run Wang, Felix Juefei-Xu, Yihao Huang, Qing Guo, Xiaofei Xie, Lei Ma, Yang Liu

Published: 2020-05-28 04:02:52+00:00

AI Summary

DeepSonar detects AI-synthesized fake voices by analyzing layer-wise neuron activation patterns of a speaker recognition system. This approach achieves high detection rates (98.1% accuracy) and low false alarm rates (around 2%) while demonstrating robustness against various manipulation attacks.

Abstract

With the recent advances in voice synthesis, AI-synthesized fake voices are indistinguishable to human ears and widely are applied to produce realistic and natural DeepFakes, exhibiting real threats to our society. However, effective and robust detectors for synthesized fake voices are still in their infancy and are not ready to fully tackle this emerging threat. In this paper, we devise a novel approach, named emph{DeepSonar}, based on monitoring neuron behaviors of speaker recognition (SR) system, ie, a deep neural network (DNN), to discern AI-synthesized fake voices. Layer-wise neuron behaviors provide an important insight to meticulously catch the differences among inputs, which are widely employed for building safety, robust, and interpretable DNNs. In this work, we leverage the power of layer-wise neuron activation patterns with a conjecture that they can capture the subtle differences between real and AI-synthesized fake voices, in providing a cleaner signal to classifiers than raw inputs. Experiments are conducted on three datasets (including commercial products from Google, Baidu, etc) containing both English and Chinese languages to corroborate the high detection rates (98.1% average accuracy) and low false alarm rates (about 2% error rate) of DeepSonar in discerning fake voices. Furthermore, extensive experimental results also demonstrate its robustness against manipulation attacks (eg, voice conversion and additive real-world noises). Our work further poses a new insight into adopting neuron behaviors for effective and robust AI aided multimedia fakes forensics as an inside-out approach instead of being motivated and swayed by various artifacts introduced in synthesizing fakes.

Key findings

DeepSonar achieves an average accuracy exceeding 98.1% and an equal error rate below 2% across three datasets. The method demonstrates robustness against manipulation attacks such as voice conversion and additive real-world noise, outperforming previous bispectral analysis methods.

Approach

DeepSonar monitors the neuron behaviors of a deep neural network (DNN)-based speaker recognition system. It uses layer-wise neuron activation patterns, determined by neuron coverage criteria (ACN and TKAN), as features for a binary classifier to distinguish real and fake voices.

Datasets

FoR (public dataset with commercial TTS voices), Sprocket-VC (self-built dataset with VC voices), MC-TTS (self-built dataset with Chinese TTS voices)

Model(s)

A third-party DNN-based speaker recognition system (using a 'thin-ResNet' architecture) and a shallow neural network binary classifier.

Author countries

Singapore, USA, China, Japan

← Previous