WaveDetect: Robust Framework for Machine-Generated Text Detection via Wavelet Transform
2026-06-22 • Computation and Language
Computation and Language
AI summaryⓘ
The authors identify problems with current methods for detecting text generated by large language models, especially when these models get better and try to hide their writing style. They introduce WaveDetect, a new way to spot machine-written text by treating the sequence of word probabilities like a signal and analyzing it using wavelet transforms to find hidden patterns. Their tests show that this method works better and is more reliable across different topics and newer models than existing detectors. This suggests that looking at text from a signal processing view can help improve detection.
Large Language ModelsText Generation DetectionAdversarial PerturbationsCross-domain ShiftsContinuous Wavelet TransformTime-Frequency DomainProbability SignalSpectral AnalysisOut-of-distribution GeneralizationMachine-generated Text
Authors
Zhichen Liu, Kaitong Qin, Linhan He, Yang Xu
Abstract
As Large Language Models asymptotically approach human-level fluency in natural language generation, solely relying on surface-level semantic artifacts for detecting LLM-generated texts has become increasingly precarious. Existing detectors often falter when facing three critical challenges: adversarial perturbations, cross-domain shifts, and the rapid temporal evolution of the foundation model. To address these issues, we propose \wavedetect, a novel framework that reformulates text detection as a signal processing task within the time-frequency domain. Unlike previous methods that analyze static token probability distributions, \wavedetect models the generated output as a probability signal, upon which a differentiable Continuous Wavelet Transform is applied to convert them into learnable spectral representations. This process reveals the intrinsic ``spectral fingerprints'' in machine-generated texts--patterns that remain invisible in time domain. Comprehensive evaluations on three well-curated datasets (RAID, EvoBench, and Domain-Shift) show that our method achieves a new state-of-the-art. It not only achieves superior accuracy but also exhibits remarkable robustness against sophisticated attacks, generalization across out-of-distribution topics and unseen evolving LLMs. Our results validate the efficacy of spectral analysis as a promising paradigm for LLM-generated texts detection.