Temporal Anchors and Editing Sensitivity in Partial Speech Spoofing: A Controlled Study

2026-10-03

AI Summary

This study addresses the false alarms and missed detections in partial speech forgery detection caused by confusion between benign edits and synthetic content. We propose a diagnostic framework based on frozen WavLM features. Methodologically, temporal anchors and boundary consistency training are introduced to analyze detector sensitivity to edits, while non-linguistic unit control and source-locking evaluation mechanisms enable forgery localization at frame, phoneme, and word granularities. Experiments reveal that phoneme-level error rates exceed those at the soft-frame level. Furthermore, data augmentation significantly reduces false alarms on authentic splices but may increase missed detections of synthetic segments, whereas consistency training yields no universal gains. This work systematically uncovers the differential impacts of augmentation strategies on detection performance.

Abstract

Partial-spoof detectors must reject synthetic content while accepting benign edits. We diagnose temporal anchors and boundary-consistency training using frozen WavLM features, nonlinguistic unit controls, and source-locked evaluation. Genuine-genuine and genuine-fake splices contrast editing false alarms with synthetic-content misses; they do not isolate a unique causal artifact. With layer-6 features, phone units have higher PartialSpoof evaluation frame EER than soft frames (11.18% versus 9.59%). In 500 retrospective PS-eval cases, augmentation reduces genuine-splice false alarms; synthetic-core misses increase at 5% PS-development FPR but not demonstrably at 1%. A 200-case Llama A follow-up retains the false-alarm reduction but does not establish a miss-rate increase. Constructed-case ranking can improve while fixed-threshold misses rise. Consistency gives no uniform gain, and encoder-layer rankings change across corpora. We report frame and event localization with three seeds, including word units, and treat PS evaluation as retrospective. Audit code, selected training scripts, and result summaries are available at https://github.com/mysxs/partial-spoof-diagnostics.

Problem

• partial speech spoofing
• spoof detection
• false alarm
• miss rate
• speech editing

Innovation

• Partial Speech Spoofing
• Temporal Anchors
• Boundary-consistency Training
• WavLM Features
• Nonlinguistic Unit Controls