openvibers/openvibe.live Commits

Public AI text rules come from openvibe-sdk/no-looks (v0.61.0, moved there verbatim); the boot repair's version follows the SDK's rules (#306)

OP OpenVibers committed 22 days ago · 141f1e2a45 · 5 files, +14 −111

Changes

ddiff --git a/README.md b/README.md
iindex 49085c3e..b309034e 100644
---- a/README.md
++++ b/README.md
@@@ -176,6 +176,6 @@ Minimum `.env`: `BASE_URL`, `JWT_SECRET`, `OV_NETWORK_URL`, `OV_NETWORK_INTERNAL
  
  <!-- versions:start -->
  - openvibe-contracts: v0.209.0
--- openvibe-sdk: v0.60.0
++- openvibe-sdk: v0.61.0
  - openvibe-shared: v4.6.0
  <!-- versions:end -->
ddiff --git a/STATUS.json b/STATUS.json
iindex f0eb7766..6e6cbc2e 100644
---- a/STATUS.json
++++ b/STATUS.json
@@@ -6,7 +6,7 @@
    "port": 3000,
    "domain": "openvibe.live (www, ingest, whip)",
    "contracts": "openvibe-contracts v0.209.0",
--  "sdk": "openvibe-sdk v0.60.0",
++  "sdk": "openvibe-sdk v0.61.0",
    "shared": "openvibe-shared v4.6.0",
    "capabilities": [
      "live.chat_context.read",
ddiff --git a/package-lock.json b/package-lock.json
iindex 445c3d91..39263e8a 100644
---- a/package-lock.json
++++ b/package-lock.json
@@@ -24,7 +24,7 @@
          "node-media-server": "^2.7.1",
          "onnxruntime-node": "^1.20.0",
          "openvibe-contracts": "https://codeload.github.com/OpenVibers/OpenVibe.Contracts/tar.gz/refs/tags/v0.209.0",
--        "openvibe-sdk": "https://codeload.github.com/OpenVibers/OpenVibe.SDK/tar.gz/refs/tags/v0.60.0",
++        "openvibe-sdk": "https://codeload.github.com/OpenVibers/OpenVibe.SDK/tar.gz/refs/tags/v0.61.0",
          "openvibe-shared": "https://codeload.github.com/OpenVibers/OpenVibe.Shared/tar.gz/refs/tags/v4.6.0",
          "pg": "^8.23.1",
          "sanitize-html": "^2.17.7",
@@@ -2660,9 +2660,9 @@
        }
      },
      "node_modules/openvibe-sdk": {
--      "version": "0.60.0",
--      "resolved": "https://codeload.github.com/OpenVibers/OpenVibe.SDK/tar.gz/refs/tags/v0.60.0",
--      "integrity": "sha512-hhgpGeEZsoGh+mGWcY22ZOXL+0vTNX+q7PtaAzxwNtr57K2XgCCHl2WqRpjHjrzyxPsjOOr9LBo3VYnmNv18bw==",
++      "version": "0.61.0",
++      "resolved": "https://codeload.github.com/OpenVibers/OpenVibe.SDK/tar.gz/refs/tags/v0.61.0",
++      "integrity": "sha512-rhQKwM+yzVjhgUPTeqMOkL9N38fbSy8JYiEbJ8SwZk8LGcusde6LoQNhQ8UUxMJIwyY7xR1Rp5NVd6XRaVgTQQ==",
        "license": "MIT",
        "bin": {
          "openvibe-download": "bin/openvibe-download.js"
ddiff --git a/package.json b/package.json
iindex 48349ba4..02a9a719 100644
---- a/package.json
++++ b/package.json
@@@ -43,7 +43,7 @@
      "node-media-server": "^2.7.1",
      "onnxruntime-node": "^1.20.0",
      "openvibe-contracts": "https://codeload.github.com/OpenVibers/OpenVibe.Contracts/tar.gz/refs/tags/v0.209.0",
--    "openvibe-sdk": "https://codeload.github.com/OpenVibers/OpenVibe.SDK/tar.gz/refs/tags/v0.60.0",
++    "openvibe-sdk": "https://codeload.github.com/OpenVibers/OpenVibe.SDK/tar.gz/refs/tags/v0.61.0",
      "openvibe-shared": "https://codeload.github.com/OpenVibers/OpenVibe.Shared/tar.gz/refs/tags/v4.6.0",
      "pg": "^8.23.1",
      "sanitize-html": "^2.17.7",
ddiff --git a/server/db/database.js b/server/db/database.js
iindex 2731ad41..61541968 100644
---- a/server/db/database.js
++++ b/server/db/database.js
@@@ -384,109 +384,12 @@ function _plainProse(text) {
          .map((l) => (/[.!?…"')\]]$/.test(l) ? l : `${l}.`))
          .join(' ');
  }
--// Public AI text never describes how a person looks (frame notes show on VOD pages and
--// feed every overview). Text written before the prompts said so ("A shirtless man sits…",
--// "a young boy at a desk") has that person called "the streamer" (or "someone" when only
--// an age is named), so a plain "a man" or "the streamer" stays as written.
--const _LOOKS = 'shirtless|topless|bare-chested|young|younger|teenage|teen|little|bearded|bald|blonde?|(?:dark|long|short|curly|red|gr[ae]y|black|brown)-haired|tattooed|muscular|heavyset|older|elderly|middle-aged|skinny|slim|chubby|attractive|pretty|handsome|cute';
--const _PEOPLE = 'man|woman|boy|girl|guy|gal|kid|child|teen|teenager|lady|gentleman|male|female|person|individual|streamer|youth';
--const _AGE_NOUN = /^(?:boy|girl|kid|child|teen|teenager|youth)$/i;
--const _LOOKED_AT = new RegExp(`\\b(a|an|the|one)\\s+((?:(?:${_LOOKS})[\\s,]+(?:and\\s+)?)*)(${_PEOPLE})\\b`, 'gi');
--const _LOOKED_AT_BARE = new RegExp(`\\b(?:(?:${_LOOKS})[\\s,]+(?:and\\s+)?)+(${_PEOPLE})\\b`, 'gi');
--const _cap = (m, s) => (/^[A-Z]/.test(m) ? s[0].toUpperCase() + s.slice(1) : s);
--// What someone wears or how their hair and face look ("wearing a black beanie and hoodie", "in a red shirt",
--// "with blond hair and a mustache", "a headset-wearing host") goes; the sentence keeps everything else. Quoted
--// speech ("Heard: "I like your hair"") is what was said and stays word for word.
--const _ART = '(?:(?:a|an|the|his|her|their|some|matching)\\s+)?';
--const _WORN = 't-shirts?|shirts?|tees?|tank tops?|hoodies?|sweatshirts?|sweaters?|jackets?|coats?|vests?|beanies?|baseball caps?|caps?|hats?|helmets?|headsets?|headphones|sunglasses|glasses|outfits?|dress(?:es)?|jerseys?|suits?|robes?|bandanas?|scarf|scarves|masks?|apparel|attire';
--const _FACE = 'beards?|mustaches?|moustaches?|goatees?|stubble|sunglasses|glasses|tattoos?|hair|ponytails?|braids|dreadlocks|piercings?|freckles|makeup|expressions?|beanies?|backward caps?|baseball caps?|hoodies?';
--const _WORN_ANY = `${_WORN}|microphones?|mics?|earbuds?|earphones?|belts?|gloves|boots|shoes|sneakers|jeans|pants|shorts|necklaces?|chains?|earrings|watch(?:es)?|uniforms?|clothes|clothing|gear|costumes?|wigs?|crowns?|goggles|visors?|lanyards?|tops?|neckties?|bow ties?|ties?|headbands?|garments?|turbans?`;
--const _WORN_NP = `${_ART}(?:[\\w'-]+\\s+){0,3}?\\b(?:${_WORN_ANY})\\b(?:\\s+(?:on|around|over|across)\\s+(?:his|her|their|the|its)\\s+(?:head|neck|face|forehead|shoulders?|ears?|eyes|chest))?(?:\\s+with\\s+(?:(?:a|an)\\s+)?(?:[\\w'-]+\\s+){0,2}?(?:zippers?|logos?|prints?|patterns?|designs?|graphics?|hoods?|stripes?|collars?|text|eyes|ears|horns|patch(?:es)?|badges?|stickers?|emblems?|lettering|writing|trim|buttons|accents|embellishments|details|sleeves))?`;
--const _WEARING = new RegExp(`,?\\s*(?:who\\s+(?:is\\s+)?|while\\s+|and\\s+)?(?<!-)\\b(?:wearing|wears|dressed in|sporting)\\s+${_WORN_NP}(?:\\s*(?:,|and|or|with|over|under)\\s+${_WORN_NP})*`, 'gi');
--const _IN_WORN = new RegExp(`\\s+in\\s+(?:(?:a|an|the|his|her|their)\\s+)?(?:[\\w'-]+\\s+){0,3}?\\b(?:${_WORN})\\b(?:\\s+(?:and|or)\\s+${_ART}(?:[\\w'-]+\\s+){0,2}?\\b(?:${_WORN_ANY})\\b)*`, 'gi');
--const _PERSON_WITH_WORN = new RegExp(`\\b((?:${_PEOPLE}|broadcaster|host|guy|gamer|player|viewer|someone|people|men|women|speaker|presenter)\\s+)with\\s+${_ART}(?:[\\w'-]+\\s+){0,3}?\\b(?:${_WORN_ANY}|${_FACE})\\b(?:\\s*(?:,|and|or)\\s+${_ART}(?:[\\w'-]+\\s+){0,2}?\\b(?:${_WORN_ANY}|${_FACE})\\b)*\\s*`, 'gi');
--const _WITH_FACE = new RegExp(`\\s+with\\s+${_ART}(?:[\\w'-]+\\s+){0,2}?\\b(?:${_FACE})\\b(?:\\s*(?:,|and|or)\\s+${_ART}(?:[\\w'-]+\\s+){0,2}?\\b(?:${_FACE}|${_WORN})\\b)*`, 'gi');
--const _WEARS_AND = new RegExp(`(?<!-)\\bwears\\s+${_WORN_NP}(?:\\s*(?:,|and|or|with|over|under)\\s+${_WORN_NP})*\\s+and\\s+(?=\\w)`, 'gi');
--// "…, while the other wears a dusty t-shirt;" goes as a whole clause.
--const _WEARS_CLAUSE = new RegExp(`,?\\s+(?:while|and|as|whereas)\\s+(?:the other|one|another|someone|he|she|they|the streamer|(?:a|the) (?:man|woman|person|guy))\\s+wears\\s+${_WORN_NP}(?:\\s*(?:,|and|or|with|over|under)\\s+${_WORN_NP})*(?=\\s*(?:[;,.!?]|$))`, 'gi');
--// "a plaid/checkered shirt": a slashed pair of words before what is worn reads as one word, so the rules below see it.
--const _SLASHED_WORN = new RegExp(`\\b([\\w'-]+)/[\\w'-]+(?=\\s+(?:[\\w'-]+\\s+){0,2}?(?:${_WORN_ANY})\\b)`, 'gi');
--// ", with a blue jacket partially visible": a comma-led clause about clothes, whoever wears them, goes (clothes only:
--// "a desk with a headset" stays).
--const _WITH_CLOTHES = /,\s*with\s+(?:(?:a|an|the|their|his|her)\s+)?(?:[\w'-]+\s+){0,2}?(?:t-shirts?|shirts?|tees?|tank tops?|hoodies?|sweatshirts?|sweaters?|jackets?|coats?|vests?|outfits?|dress(?:es)?|jerseys?|suits?|robes?|outerwear)\b[^,.;!?]*(?:,(?=\s+(?:is|are|was|were)\s))?/gi;
--function _noClothing(text) {
--    return text.split(/("[^"]*"?|“[^”]*”?)/).map((part, i) => (i % 2 ? part : part
--        .replace(_SLASHED_WORN, '$1')
--        .replace(_WITH_CLOTHES, '')
--        // A parenthetical clause taken out of the middle ("A third person, with a red cap, is shown") leaves no ",,".
--        .replace(/\s*,\s*,\s*/g, ' ')
--        .replace(/\b[\w-]+-wearing\s+/gi, '')
--        .replace(_WEARS_AND, '')
--        .replace(_WEARS_CLAUSE, '')
--        .replace(/\b(?:casually|smartly|neatly|warmly|brightly|formally|well)[\s-]+dressed\s+/gi, '')
--        .replace(_PERSON_WITH_WORN, '$1')
--        .replace(_WEARING, '')
--        .replace(_IN_WORN, '')
--        .replace(_WITH_FACE, '')
--        .replace(/\s+([,.;:!?])/g, '$1')
--        .replace(/,(?=[.;:!?])/g, '')
--        .replace(/([:;]),/g, '$1')
--        .replace(/ {2,}/g, ' '))).join('');
--}
--// Tags shown with public AI items ("red-beanie", "bearded", "man", "tanktop") follow the same rule: a tag naming
--// clothing, hair, face, age or a person by gender is dropped.
--const _LOOKS_TAG = new RegExp(`(?:^|[\\s_-])(?:${_WORN}|${_FACE}|${_LOOKS}|men|women|man|woman|boy|girl|guy|gal|kid|child|teen|teenager|lady|male|female|youth|shirtless|topless|tanktop|selfie)(?=$|[\\s_-])`, 'i');
--// A short AI title that names looks ("Orange Shirt Kitchen Chat") cannot be repaired word by word: it is not used.
--const _LOOKS_WORD = new RegExp(`\\b(?:${_WORN}|${_FACE}|${_LOOKS}|faces?|smil(?:e|ing)|body|skin|men|women|man|woman|boy|girl|guy|kid|child|teen|lady|male|female|shirtless|topless)\\b`, 'i');
--const _hasLooks = (t) => _LOOKS_WORD.test(String(t || ''));
--function _noLooksTags(tags) {
--    return (Array.isArray(tags) ? tags : []).filter((t) => typeof t === 'string' && t.trim() && !_LOOKS_TAG.test(t.trim()));
--}
--// Body language is the person too ("reacting with a thoughtful pose", "in a relaxed posture"): the phrase goes.
--const _POSE = /,?\s*(?:(?:appearing|seeming|seemingly)\s+(?:to\s+be\s+)?)?(?:(?:reacting|sitting|standing|leaning|posing|looking|resting)\s+)?(?:with|in)\s+(?:an?|their|his|her)\s+(?:[\w-]+\s+){0,2}(?:pose|posture|expression|facial expression|stance)\b/gi;
--// A guessed gender is a description of the person too (owner's rule: no looks, body, clothing, hair, age or guessed
--// gender): "a man in a kitchen" is "a person in a kitchen", "two men" are "two people", "a male or female host" is "a
--// host". Only after an article or a count, so names (Iron Man, Pac-Man, Spice Girls) and quoted speech stay.
--const _GENDERED = 'man|woman|guy|gal|lady|gentleman';
--const _GENDERED_ONE = new RegExp(`\\b(a|an|the|one|another|this|that|each|every)\\s+(?:${_GENDERED})\\b(?!-)`, 'gi');
--const _GENDERED_MANY = new RegExp(`\\b(two|three|four|five|several|some|the|both|many|other|few|these|those|\\d+)\\s+(?:men|women|guys|gals|ladies|gentlemen)\\b`, 'gi');
--const _GENDER_ADJ = /\b(?:(?:a|an)\s+)?(?:male|female)\s+(?:or|and|\/)\s+(?:male|female)\s+|\b(?:male|female)\s+(?=(?:hosts?|streamers?|players?|speakers?|presenters?|gamers?|voices?|narrators?|commentators?|figures?|characters?|broadcasters?|viewers?|individuals?|persons?|people)\b)/gi;
--// What is left of a sentence about someone's hair, clothes or body ("a person adjusts their hair or clothing near the
--// counter") cannot be repaired word by word: the sentence goes.
--const _GROOMING = new RegExp('\\b(?:(?:their|his|her)\\s+(?:own\\s+)?(?:hair|beard|face|makeup|clothing|clothes|outfit|appearance|body|figure|looks)'
--    // The face and body described ("smiling broadly, displaying white teeth and a beard") and clothes the word lists do
--    // not name ("blue outerwear", "in outdoor gear"): the sentence goes. Things a person holds or reviews (headphones, a
--    // headset on a desk) are not here.
--    + '|beards?|mustaches?|moustaches?|teeth|grin(?:s|ning)?|smil(?:e|es|ing)\\s+(?:broadly|widely|warmly|at)|freckles|complexion|wrinkles'
--    + '|hairstyles?|haircuts?|outerwear|(?:in|wearing)\\s+(?:[\\w-]+\\s+){0,2}gear|face\\s+shields?)\\b', 'i');
--function _noGender(text) {
--    return text.split(/("[^"]*"?|“[^”]*”?)/).map((part, i) => (i % 2 ? part : part
--        .replace(_GENDER_ADJ, (m) => (/^an?\s/i.test(m) ? m.slice(0, m.indexOf(' ') + 1) : ''))
--        .replace(_GENDERED_ONE, (m, art) => _cap(m, `${/^an$/i.test(art) ? 'a' : art.toLowerCase()} person`))
--        .replace(_GENDERED_MANY, (m, n) => _cap(m, `${n.toLowerCase()} people`))
--        .replace(/\ba\s+(?=[aeiou]\w*\s+(?:host|streamer|player|speaker|presenter|gamer|figure|character|broadcaster|person)\b)/gi, (m) => _cap(m, 'an '))
--        .replace(/ {2,}/g, ' '))).join('');
--}
--function _noGrooming(text) {
--    const sentences = text.match(/[^.!?]+(?:[.!?]+(?:["”])?|$)\s*/g) || [text];
--    return sentences.filter((s) => !_GROOMING.test(s.replace(/"[^"]*"?|“[^”]*”?/g, ''))).join('').trim();
--}
--function _noLooks(text) {
--    if (!text || typeof text !== 'string') return text;
--    return _noGrooming(_noGender(_noLooksWords(text)));
--}
--function _noLooksWords(text) {
--    return _noClothing(text).replace(_POSE, '').replace(/\s+([,.;:!?])/g, '$1').replace(/ {2,}/g, ' ')
--        // Looks named: the person on camera. Only an age ("a child sits nearby"): someone, who may not be the streamer.
--        .replace(_LOOKED_AT, (m, art, looks, noun) => (looks ? _cap(m, 'the streamer') : _AGE_NOUN.test(noun) ? _cap(m, /^the$/i.test(art) ? 'the person' : 'someone') : m))
--        .replace(_LOOKED_AT_BARE, (m) => _cap(m, 'streamer'))
--        .replace(/\s+(?:is|are)\s+(?:shirtless|topless|bare-chested)\s+and\s+/gi, ' ')
--        .replace(/\b(?:shirtless|topless|bare-chested)\s+and\s+/gi, '')
--        .replace(/\b(one|the other|another),?\s+and\s+(?=\w+ing\b)/gi, '$1 ')
--        .replace(/\s+(?:is|are)\s+(?:shirtless|topless|bare-chested)\b/gi, ' is on camera')
--        .replace(/\s*,\s*(?:shirtless|topless|bare-chested)\s*,/gi, '')
--        .replace(/\b(?:shirtless|topless|bare-chested)\b,?\s*/gi, '');
--}
++// Public AI text never describes how a person looks (the owner's rule: no looks, body, clothing, hair, face, pose,
++// age or guessed gender). Frame notes show on VOD pages and feed every overview, so every public AI string passes
++// noLooks() (tags noLooksTags(); a short title that names looks is replaced, hasLooks()). The rules and their test
++// corpus are openvibe-sdk/no-looks since SDK 0.61.0 (moved from here, Live#298–#302), shared with Community and
++// MediaHub; test/ai-text-no-looks.test.js still checks Live's writers and the repair below.
++const { noLooks: _noLooks, noLooksTags: _noLooksTags, hasLooks: _hasLooks, RULES_VERSION: _LOOKS_RULES } = require('openvibe-sdk/no-looks');
  /**
   * A stream memory as the public sees it (home "Right now", a VOD's AI timeline, a streamer's AI profile): the
   * observation only. The " — heard: "…"" clause is raw speech-to-text kept for the AI's own context (often another
@@@ -749,7 +652,7 @@ let _looksRulesVersion = null;
  // What the repair covers is part of its version: a change in scope runs the scan again even when the rules did not change.
  const _LOOKS_REPAIR_SCOPE = 'full and short columns, each on its own (2026-10-22)';
  function looksRulesVersion() {
--    if (!_looksRulesVersion) _looksRulesVersion = require('crypto').createHash('sha1').update([_LOOKS_REPAIR_SCOPE, _WITH_CLOTHES, 'no double commas; earbud', _noLooks, _noLooksWords, _noGender, _noGrooming, _noClothing, _LOOKS, _PEOPLE, _WORN, _WORN_ANY, _FACE, _GENDER_ADJ, _GROOMING].map(String).join('|')).digest('hex').slice(0, 12);
++    if (!_looksRulesVersion) _looksRulesVersion = require('crypto').createHash('sha1').update([_LOOKS_REPAIR_SCOPE, _LOOKS_RULES].join('|')).digest('hex').slice(0, 12);
      return _looksRulesVersion;
  }
  async function cleanupMalformedAiText({ force = false } = {}) {