用可变时间常数网络检测语音深度伪造，小模型实现高精度跨数据集识别。
FlowFake: Liquid Networks for Audio Deepfake Detection
- 采用自适应时间常数的微分方程架构，同步捕捉短时频谱与长时韵律特征。
- 仅34K参数，在4个数据集上跨域检测准确率最高达79.97%。
- 适合需要轻量级、强泛化能力的语音安全场景应用。
由神经文本转语音和语音克隆系统生成的语音深度伪造对说话人验证和公共舆论构成大规模威胁。核心挑战在于跨数据集泛化：在一种合成流程上训练的检测器在未见过的伪造样本上性能崩溃。我们指出，这一失败主要源于多时间尺度的结构化合成语音异常。现有检测器均采用固定窗口帧统计，其架构与信号特性不匹配。本文提出FlowFake，一种基于液态时间常数（LTC）的架构，其隐藏状态通过学习的常微分方程演化，每个神经元拥有自适应时间常数，同时捕捉10ms频谱与2s韵律线索。该模型仅需34K参数，具备正式的有界输入有界输出（BIBO）稳定性及O(dt^4)积分误差。在四个数据集的跨域基准测试中（ASVspoof2019-LA、FakeOrReal、InTheWild、MLAAD），仅在FakeOrReal上训练即达到75.29%准确率，仅在MLAAD上训练达79.97%。优于RawGAT-ST与Whisper-DF所有对比对，且以0.01%的参数量媲美300倍大的SSL Wav2vec2。源代码已公开于https://github.com/GhostRider2023/FlowFake。
原文摘要 · Abstract (English)
Audio deepfakes generated by neural text-to-speech and voice-cloning systems threaten speaker verification and public discourse at scale. The core challenge is cross-dataset generalization: detectors trained on one synthesis pipeline collapse on unseen forgeries. We argue that this failure is primarily because of structural synthetic speech artifacts which are multi-timescale trajectory anomalies. Though every existing detector aggregates a fixed-window frame statistics, this misaligns the architecture with the signal. We propose FlowFake, a Liquid Time-Constant (LTC) architecture whose hidden state evolves via a learned ODE, with per-neuron adaptive time constants simultaneously resolving spectral (10ms) and prosodic (2s) cues. At only 34K parameters FlowFake achieves formal BIBO stability and O(dt^4) integration error. On a four-dataset cross domain benchmark (ASVspoof2019-LA, FakeOrReal, InTheWild, MLAAD), FlowFake reaches 75.29% on ASVspoof2019 trained only on FakeOrReal and 79.97% trained only on MLAAD. It outperforms RawGAT-ST and Whisper-DF on every evaluated pair and matching SSL Wav2vec2 (300x larger) at 0.01% of its parameter count. The source code is available on : https://github.com/GhostRider2023/FlowFake
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。