RTCFake: Speech Deepfake Detection in Real-Time Communication

Jun Xue, Zhuolin Yi, Yihuan Huang, Yanzhen Ren, Yujie Chen, Cunhang Fan, Zicheng Su, Yongcheng Zhang, Bo Cai

Overview

The RTCFake dataset is the first large-scale speech deepfake dataset tailored for real-time communication scenarios, which contains approximately 600 hours of speech. The dataset is constructed by transmitting speech through multiple mainstream social media and conferencing platforms (e.g., Zoom), enabling precise pairing between offline and online speech. This dataset captures the complex, nonlinear distortions introduced by real-world "black-box" transmission, such as unknown noise suppression, echo cancellation and codec compression, providing a more realistic and challenging evaluation benchmark for speech deepfake detection.

Demos

Offline Speech

Training Subset

	Real	F5-TTS	OpenAudio-S1	VOXCPM	LLaSA	CosyVoice	SeedVC
Chinese						-	-
English					-

* The missing entries in the table arise because the methods used to generate the Chinese and English data are not entirely identical.

Development Subset

	Real	F5-TTS	OpenAudio-S1	VOXCPM	LLaSA	CosyVoice	SeedVC
Chinese						-	-
English					-

Evaluation Subset

	Real	F5-TTS	OpenAudio-S1	VOXCPM	LLaSA	IndexTTS2	Doubao	SparkTTS	CosyVoice	SeedVC	ChatterboxVC
Chinese							-	-
English

Paired Offline-online Speech

Comparison of Identical Real and Fake Utterances Across Multiple Transmission Platforms

	Offline	QQ	Zoom	DingTalk	Voov	Lark	WeChat	Telegram
Real
Fake (IndexTTS2)

Comparison of Identical Clean and Noisy Utterances Across Multiple Transmission Platforms

	Offline	QQ	Zoom	DingTalk	Voov	Lark	WeChat	Telegram
Clean
Noisy (Rain)

Utterances with Different Noises Transmitted through the Same Platform (Lark)

	Noise_keyboard	Noise_footsteps	Noise_rain	Noise_coffee	Noise_office	Echo
Offline
Online (Lark)