Quantum Kernel Q-Patch Improves Audio Deepfake Detection
A new method called Q-Patch has been introduced by researchers for detecting audio deepfakes. This quantum feature map transforms local time-frequency patches from mel-spectrograms into quantum states. It utilizes shallow circuits that are efficient in hardware and incorporate adjacency-aware entanglement. Each patch is represented by a four-dimensional acoustic descriptor and is linked to a four-qubit circuit with a maximum depth of three. In tests conducted on an audio spoofing detection task using a controlled and balanced protocol, Q-Patch surpassed classical baselines of similar size, achieving a superior area under the receiver operating characteristic curve. This technique specifically leverages the time-frequency characteristics of spectrograms rather than treating them as standard images.
Key facts
- Q-Patch is a quantum feature map tailored to audio
- Encodes local time-frequency patches from mel-spectrograms into quantum states
- Uses shallow, hardware-efficient circuits with adjacency-aware entanglement
- Each patch is summarized by a compact four-dimensional acoustic descriptor
- Mapped to a four-qubit circuit with depth at most three
- Evaluated on audio spoofing detection task with controlled, balanced protocol
- Outperforms size-matched classical baselines
- Achieves higher area under the receiver operating characteristic curve
Entities
—