setec-voiceprint

Text-only stylometry for authorial voice in prose — not audio voice identification or speaker recognition. Stylometric tools for AI-prose smoothing diagnosis and authorial-voice coherence comparison. Five task surfaces: smoothing diagnosis (Layer A variance signals on prose), authorial-voice coherence comparison (Burrows-style Delta + cluster mode + idiolect preservation lists + authorial-voice drift tracking across time periods + per-POV authorial-voiceprints with voice-collapse detection + impostor-pool acquisition for the General Imposters method), validation + calibration (manifest integrity + corpus-hygiene gate + ROC/PR harness on smoothing and authorial-voice coherence + voice-validation harness + per-comparator + per-(judge × generator) direction-routed calibration toolchain with bake-off matrix runner + length-stratified subsampling + slicer/polarity-audit chainer + provenance ledger), craft-restoration (AIC flag families, source triage, rhetorical countermoves, metric-targeted restoration packets that translate diagnostic outputs into bounded revision-safe prompts, before/after restoration verdicts with metric-gaming detection), and discrimination evidence shipped uncalibrated by default (Hans et al. 2024 Binoculars two-model perplexity audit + the SETEC external-mirror multi-LLM continuation-distance evidence pack). Python tooling spans manuscript-level audits, sliding-window localization, length-matched bootstrap percentiles, R `stylo` oracle verification, MVP empirical validation against labeled corpora, live blog acquisition (Substack / WordPress / generic-HTML / Wayback), offline Blogger Takeout import, online literary-horror magazine scraping (Nightmare, The Dark, with per-magazine modules behind a uniform CLI), and PDF library inventory + extraction (pypdf text-layer + optional ocrmypdf OCR) for the impostor pool. 'Voiceprint' in the name refers to authorial voice in prose (Burrows-Delta-style authorship fingerprinting), not vocal spectography.

8 skills