Armiger Plugin #4 — Cross-Head Audio Language & Anime Dub Guard (armiger-audio-guard)

Couples Armiger's neural title_language and lang heads with linguistic & tracker-tag disambiguation to lift binary English-vs-Foreign audio language accuracy on oct15_confirmation (N=545) from 96.88% to 97.80% (92.75% 3-class, N=772).
Available standalone here and pre-bundled inside NorviTech/armiger-ultra.

Bundled in Armiger Ultra Compatible with Armiger Licence Apache-2.0

What armiger-audio-guard Solves

Standard regex parsers (RTN, Parsett, GuessIt, PTN) and single-head classifiers systematically fail on five recurring multilingual torrent patterns:

  1. Romance-Language .en. Prepositions: Titles like Le.Voyage.dans.la.Lune.1902.1080p.mp4 or Les.Quatre.Cents.Farces.du.Diable.1906.mkv contain the Spanish/French word en ("in"). Case-insensitive .en. tokenizers misclassify these as English audio.
  2. <Foreign>.ENG.Sub Releases: In Movie.Title.ITA.ENG.Sub.2025.mkv, ENG modifies Sub (Italian audio with English subtitles), not a dual-audio track.
  3. Regional Dub Tags without Explicit Language Cues: Releases tagged Hindi Dubbed, German.AC3LD, Dual-Lat (Latin American Spanish), BG Audio (Bulgarian), or Dutch network (BNN) ... NL.
  4. Parenthesized / Bracketed Multi-Audio Lists: MULTi (FR EN) or [RUS + JAP + ENG] where English is listed inside a multi-audio group.
  5. Untagged Foreign-Language Releases with English Technical Boilerplate: Foreign-language films uploaded with English encoding tags (1080p.WEBM.x264) where Armiger's neural title_language head (P(non_english_title)) detects the non-English title stem.

Measured Accuracy (Unmutated Ground-Truth Keys)

Benchmark Cohort (Unmutated rows.y) GuessIt RTN / Parsett Standalone Neural lang Head With armiger-audio-guard (Armiger / Ultra)
oct15_confirmation Binary EN-vs-Foreign (N=545) 96.01% 95.33% 96.88% 97.80%
oct15_confirmation 3-Class Audio (N=772) — — 68.39% (raw prior) 92.75% (calibrated + guard)
multilang_3000 Audio (N=3,000) — — 91.47% 91.47%
oct15_confirmation Anime Detection (N=507) 75.90% 77.65% 99.61% 99.61% (505/507)
zero_cue_2000 Anime Detection (N=1,998) 75.90% 77.65% 95.80% 95.80% (1,914/1,998)

Quickstart

from audio_language_guard import apply_audio_language_guard, classify_audio_and_anime_dub

# Example 1: Spanish preposition '.en.' with high neural foreign-title probability
p_en = apply_audio_language_guard(
    raw_name="Le.Voyage.dans.la.Lune.1902.1080p-dual-lat.mp4",
    p_en=0.62,
    p_title_foreign=0.96,
)
print("Adjusted P(English audio):", p_en)  # -> 0.10 (Foreign audio)

# Example 2: Multi-audio release with explicit (FR EN)
print(classify_audio_and_anime_dub(
    raw_name="Show.Name.S01.MULTi (FR EN) [1080p]",
    p_en=0.38,
    p_anime=0.02,
    p_title_foreign=0.15,
))

Part of the Armiger Modular Suite

This plugin is bundled inside NorviTech/armiger-ultra alongside all 6 official plugins (armiger-swarm-survival, armiger-swarm-priors, armiger-zero-cue-catalog, armiger-audio-guard, armiger-calibration-dht, and armiger-dht-db-sync) and is compatible with NorviTech/armiger (8.3 MB CPU INT8).

Intended Use & Legal Compliance

Open-source machine learning and cryptographic metadata-classification utility designed strictly for lawful use cases (indexing licensed, public-domain, Creative Commons, and open-source media/software; malware & fake-seeder honeypot filtering; and CSAM/abuse pre-ingest blocking). NorviTech does not condone, encourage, or support copyright infringement. See LEGAL.md and NOTICE.


Part of the NorviTech Suite — open-source apps for the Linux desktop and the self-hosted stack

Armiger Ultra · Armiger · Swarm Survival · Swarm Priors · Zero-Cue Catalog · Audio Guard · Calibration & DHT RSI · DHT DB Sync · norvitech.com

Downloads last month
31
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collection including NorviTech/armiger-audio-guard