Compact screen spots fake speech using sound signature traces
Tracing Decoder Artifacts for Compact Synthetic Speech Screening
SoundArtificial Intelligence
Summary
Detecting synthetic or fake speech usually needs big, slow computer models. The authors designed a small, fast tool that looks for tiny predictable sound patterns left behind by the way fake speech is made. This tool acts like a cheap first check, only passing suspicious audio to the big models for a detailed check. It can detect many kinds of synthetic voices accurately and saves a lot of computing power by working as a filter before bigger detectors.
What this means in practice
- •For security teams: Screen incoming audio for fake speech quickly to filter suspicious recordings before expensive deep analysis.
- •For voice assistant developers: Reduce compute costs by adding a fast pre-check for synthetic speech to improve detection pipelines without slowing response times.
Authors
Yi Chen Liu, Jian Liu
Abstract
Recent advances in speech synthesis and voice cloning have increased the need for reliable synthetic-speech detection, yet high-accuracy detectors increasingly rely on large pretrained models that are costly to invoke on every recording. Rather than replacing such detectors, we investigate a compact front-end screen that processes all inputs cheaply and forwards only suspicious recordings for more expensive analysis. To enable lightweight screening without a large learned encoder, we exploit spectral traces introduced by speech-generation operations. We analyze how learned upsampling and inverse short-time Fourier transform synthesis can produce predictable spectral artifacts and measure their presence directly in generated waveforms. Because the strength of these artifacts varies across generators, we combine decoder-guided spectral measurements with complementary descriptors of short-time spectral shape and temporal variation in a compact gradient-boosted tree. Across seven speech generators and two human-speech sources, the proposed screen achieves an equal error rate of 0.021\% with an estimated model storage of 151 KiB. When used as the first stage of a simulated cascade with a 1.15-billion-parameter detector, it reduces estimated detection energy by 84.4\% while operating at a 0.050\% synthetic-speech miss rate, demonstrating the potential of decoder-guided acoustic evidence for low-cost front-end screening.