Metacrowd DeepForensics

DeepForensics is a deepfake detection and forensic analysis service that examines audio and video for signs of forgery or manipulation. It combines three detection engines and cross-checks their results. Detection engines: - DST-Net: deep voice detection engine that evaluates audio for AI-synthesized speech and voice conversion, with risk assessed per audio segment. - CREST-Net: video deepfake detection engine that analyzes frame and temporal visual features to assess face synthesis or swapping and partial manipulation. - Multimodal detection engine: analyzes audio-video synchronization and person/speech consistency to find inconsistencies between modalities. Workflow: - Input of audio/video files and basic metadata. - Modality separation and preprocessing of audio, frames, and timeline information. - Independent analysis per engine, followed by multimodal cross-verification. - Alignment of suspicious segments on a common timeline. - A forensic report with overall risk, suspicious segments, per-modality evidence, and review points. Results are kept separate per engine and structured by segment, modality, and evidence rather than a single probability score. When an original file is available, the workflow can be extended to compare original and suspect content. The vendor states the results are reference material for technical review and do not replace legal forensic appraisal. Listed use cases: digital forensics and case investigation, media and news authenticity verification, copyright and originality analysis, and platform content verification (UGC and reported content).