Back to browse
GitHub Repository
0 starsPython

Real-Life Deception Detection Without Uploading Video

by ksingla025·Jun 21, 2026·2 points·2 comments

AI Analysis

MidBold Bet

Requires external Whissle gateway with auth token; core ML runs on their service, not locally.

Strengths
  • Leave-One-Speaker-Out cross-validation prevents models from memorizing speaker identity instead of deception cues.
  • Three-lane architecture fuses transcript metadata, visual markers, and local librosa prosody features.
Weaknesses
  • External gateway dependency means you cannot run the pipeline without their proprietary service and API token.
  • Deception detection from video lacks scientific consensus; small dataset of 121 clips limits generalization claims.
Category
Target Audience

Researchers studying deception detection or multi-modal ML pipelines

Similar To

Affectiva · nVISO · iMotions

Similar Projects