import Foundation public final class TextProcessor { public static let shared = TextProcessor() private static let trailingEndRegex: NSRegularExpression = { return try! NSRegularExpression(pattern: #"[،,:\s]+(and|تمام|end)[.!?]?$"#, options: .caseInsensitive) }() private static let singleEndRegex: NSRegularExpression = { return try! NSRegularExpression(pattern: #"^(and|تمام|end)[.!?]?$"#, options: .caseInsensitive) }() private init() {} public func process(text: String) -> String { var result = text.trimmingCharacters(in: .whitespacesAndNewlines) if result.isEmpty { return "" } // 0. Check for Hermes / Bot Messenger Commands (/steer, /queue, /cancel) if let botCmd = BotCommandProcessor.parse(text: result) { let processedBody: String if !botCmd.body.isEmpty { var bodyText = botCmd.body if SonioxSettings.shared.voicePunctuation { bodyText = applyVoicePunctuation(bodyText) } if SonioxSettings.shared.domainVocabulary { bodyText = applyDomainVocabulary(bodyText) } bodyText = applyNumberFormatting(bodyText, format: SonioxSettings.shared.numberFormat) if SonioxSettings.shared.normalizeText { bodyText = normalizeText(bodyText) } processedBody = bodyText.trimmingCharacters(in: .whitespacesAndNewlines) } else { processedBody = "" } if processedBody.isEmpty { return botCmd.command } else { return "\(botCmd.command)\n\(processedBody)" } } // 1. Voice Punctuation Commands if SonioxSettings.shared.voicePunctuation { result = applyVoicePunctuation(result) } // 2. Domain Vocabulary Enhancement if SonioxSettings.shared.domainVocabulary { result = applyDomainVocabulary(result) } // 3. Number Formatting result = applyNumberFormatting(result, format: SonioxSettings.shared.numberFormat) // 3.5. Sanitize foreign hallucinations & word loops result = SonioxLiveSession.sanitizeText(result) // 4. Text Normalization if SonioxSettings.shared.normalizeText { result = normalizeText(result) } // 5. Wake Word & End Trigger sanitization result = stripTrailingEndTrigger(result) return result.trimmingCharacters(in: .whitespacesAndNewlines) } /// Lightweight inline formatter applied to live streaming chunks so words, domain terms, /// and numbers match the final post-processed text in real time. public func processLiveChunk(_ text: String) -> String { if text.isEmpty { return "" } let hasLeadingSpace = text.first?.isWhitespace == true let hasTrailingSpace = text.last?.isWhitespace == true var result = text if SonioxSettings.shared.voicePunctuation { result = applyVoicePunctuation(result) } if SonioxSettings.shared.domainVocabulary { result = applyDomainVocabulary(result) } result = applyNumberFormatting(result, format: SonioxSettings.shared.numberFormat) let trimmed = result.trimmingCharacters(in: .whitespaces) if trimmed.isEmpty { return text } return (hasLeadingSpace ? " " : "") + trimmed + (hasTrailingSpace ? " " : "") } public func applyDomainVocabularyInline(_ text: String) -> String { return applyDomainVocabulary(text) } private func applyVoicePunctuation(_ text: String) -> String { var t = " " + text + " " // Persian Voice Commands let persianMappings: [(String, String)] = [ (" علامت سوال", "؟"), (" علامت سؤال", "؟"), (" علامت پرسش", "؟"), (" علامت تعجب", "!"), (" نقطه ویرگول", "؛"), (" خط بعد", "\n"), (" سطر بعد", "\n"), (" سر خط", "\n"), (" اینتر", "\n"), (" برو خط بعد", "\n"), (" نقطه", "."), (" ویرگول", "،"), (" کاما", "،"), (" دو نقطه", ":"), (" سه نقطه", "..."), (" گیومه باز", " «"), (" گیومه بسته", "» "), (" پرانتز باز", " ("), (" پرانتز بسته", ") ") ] for (cmd, rep) in persianMappings { t = t.replacingOccurrences(of: cmd, with: rep, options: .caseInsensitive) } // English Voice Commands let englishMappings: [(String, String)] = [ (" question mark", "?"), (" exclamation mark", "!"), (" exclamation point", "!"), (" semi colon", ";"), (" semicolon", ";"), (" new line", "\n"), (" enter key", "\n"), (" full stop", "."), (" period", "."), (" comma", ","), (" colon", ":"), (" ellipsis", "..."), (" open quote", " \""), (" close quote", "\" "), (" open parenthesis", " ("), (" close parenthesis", ") ") ] for (cmd, rep) in englishMappings { t = t.replacingOccurrences(of: cmd, with: rep, options: .caseInsensitive) } return t } private static let domainTerms: [(String, String)] = [ ("این اپ پرچس", "In-App Purchase"), ("این اپ پرشرز", "In-App Purchase"), ("اپ پرشرز", "In-App Purchases"), ("اپل کانکت", "App Store Connect"), ("اپ استور کانکت", "App Store Connect"), ("حجزی مرچنت", "Hajzee Merchant"), ("حجزی یوزر", "Hajzee User"), ("حجزی", "Hajzee"), ("نیوهورایزن", "NewHorizon"), ("نیو هورایزن", "NewHorizon"), ("نیوهوریزن", "NewHorizon"), ("زر محک", "ZarMahak"), ("زرمحک", "ZarMahak"), ("سانی اوکس", "Soniox"), ("سانیوکس", "Soniox"), ("سانی اکس", "Soniox"), ("تیم بای", "TeamBy"), ("تیمبی", "TeamBy"), ("دوودی", "Dovodi"), ("داودی", "Dovodi"), ("داکر", "Docker"), ("کوبرنتیز", "Kubernetes"), ("کوبرنتس", "Kubernetes"), ("ای اس ایکس آی", "ESXi"), ("ای اس ایکس ای", "ESXi"), ("فست ای پی آی", "FastAPI"), ("فست ای پی ای", "FastAPI"), ("پستگرس کیو ال", "PostgreSQL"), ("پستگرس", "PostgreSQL"), ("ردیس", "Redis"), ("فلاتر", "Flutter"), ("وایارگارد", "WireGuard"), ("وایرگارد", "WireGuard"), ("میکروتیک", "MikroTik"), ("هرمس", "Hermes"), ("نکست جی اس", "Next.js"), ("ری اکت", "React"), ("پایتون", "Python"), ("سویفت", "Swift") ] private func applyDomainVocabulary(_ text: String) -> String { var t = text for (target, canonical) in TextProcessor.domainTerms { t = t.replacingOccurrences(of: target, with: canonical, options: .caseInsensitive) } return t } private func applyNumberFormatting(_ text: String, format: NumberFormatOption) -> String { switch format { case .raw: return text case .digits, .persianDigits: var t = " " + text + " " let compounds: [(String, String)] = [ (" بیست و نه ", " 29 "), (" بیست و هشت ", " 28 "), (" بیست و هفت ", " 27 "), (" بیست و شش ", " 26 "), (" بیست و پنج ", " 25 "), (" بیست و چهار ", " 24 "), (" بیست و سه ", " 23 "), (" بیست و دو ", " 22 "), (" بیست و یک ", " 21 "), (" سی و نه ", " 39 "), (" سی و هشت ", " 38 "), (" سی و هفت ", " 37 "), (" سی و شش ", " 36 "), (" سی و پنج ", " 35 "), (" سی و چهار ", " 34 "), (" سی و سه ", " 33 "), (" سی و دو ", " 32 "), (" سی و یک ", " 31 "), (" چهل و نه ", " 49 "), (" چهل و هشت ", " 48 "), (" چهل و هفت ", " 47 "), (" چهل و شش ", " 46 "), (" چهل و پنج ", " 45 "), (" چهل و چهار ", " 44 "), (" چهل و سه ", " 43 "), (" چهل و دو ", " 42 "), (" چهل و یک ", " 41 "), (" پنجاه و نه ", " 59 "), (" پنجاه و هشت ", " 58 "), (" پنجاه و هفت ", " 57 "), (" پنجاه و شش ", " 56 "), (" پنجاه و پنج ", " 55 "), (" پنجاه و چهار ", " 54 "), (" پنجاه و سه ", " 53 "), (" پنجاه و دو ", " 52 "), (" پنجاه و یک ", " 51 "), (" شصت و نه ", " 69 "), (" شصت و هشت ", " 68 "), (" شصت و هفت ", " 67 "), (" شصت و شش ", " 66 "), (" شصت و پنج ", " 65 "), (" شصت و چهار ", " 64 "), (" شصت و سه ", " 63 "), (" شصت و دو ", " 62 "), (" شصت و یک ", " 61 "), (" هفتاد و نه ", " 79 "), (" هفتاد و هشت ", " 78 "), (" هفتاد و هفت ", " 77 "), (" هفتاد و شش ", " 76 "), (" هفتاد و پنج ", " 75 "), (" هفتاد و چهار ", " 74 "), (" هفتاد و سه ", " 73 "), (" هفتاد و دو ", " 72 "), (" هفتاد و یک ", " 71 "), (" هشتاد و نه ", " 89 "), (" هشتاد و هشت ", " 88 "), (" هشتاد و هفت ", " 87 "), (" هشتاد و شش ", " 86 "), (" هشتاد و پنج ", " 85 "), (" هشتاد و چهار ", " 84 "), (" هشتاد و سه ", " 83 "), (" هشتاد و دو ", " 82 "), (" هشتاد و یک ", " 81 "), (" نود و نه ", " 99 "), (" نود و هشت ", " 98 "), (" نود و هفت ", " 97 "), (" نود و شش ", " 96 "), (" نود و پنج ", " 95 "), (" نود و چهار ", " 94 "), (" نود و سه ", " 93 "), (" نود و دو ", " 92 "), (" نود و یک ", " 91 ") ] for (word, num) in compounds { t = t.replacingOccurrences(of: word, with: num) } let simples: [(String, String)] = [ (" ده ", " 10 "), (" یازده ", " 11 "), (" دوازده ", " 12 "), (" سیزده ", " 13 "), (" چهارده ", " 14 "), (" پانزده ", " 15 "), (" شانزده ", " 16 "), (" هفده ", " 17 "), (" هجده ", " 18 "), (" نوزده ", " 19 "), (" بیست ", " 20 "), (" سی ", " 30 "), (" چهل ", " 40 "), (" پنجاه ", " 50 "), (" شصت ", " 60 "), (" هفتاد ", " 70 "), (" هشتاد ", " 80 "), (" نود ", " 90 "), (" صد ", " 100 "), (" دویست ", " 200 "), (" سیصد ", " 300 "), (" چهارصد ", " 400 "), (" پانصد ", " 500 "), (" ششصد ", " 600 "), (" هفتصد ", " 700 "), (" هشتصد ", " 800 "), (" نهصد ", " 900 "), (" صفر ", " 0 "), (" یک ", " 1 "), (" دو ", " 2 "), (" سه ", " 3 "), (" چهار ", " 4 "), (" پنج ", " 5 "), (" شش ", " 6 "), (" هفت ", " 7 "), (" هشت ", " 8 "), (" نه ", " 9 ") ] for (word, num) in simples { t = t.replacingOccurrences(of: word, with: num) } if format == .persianDigits { let pDigits = ["0":"۰", "1":"۱", "2":"۲", "3":"۳", "4":"۴", "5":"۵", "6":"۶", "7":"۷", "8":"۸", "9":"۹"] for (en, fa) in pDigits { t = t.replacingOccurrences(of: en, with: fa) } } return t.trimmingCharacters(in: .whitespacesAndNewlines) case .words: var t = text let toWords = [ "0": "صفر", "1": "یک", "2": "دو", "3": "سه", "4": "چهار", "5": "پنج", "6": "شش", "7": "هفت", "8": "هشت", "9": "نه", "۰": "صفر", "۱": "یک", "۲": "دو", "۳": "سه", "۴": "چهار", "۵": "پنج", "۶": "شش", "۷": "هفت", "۸": "هشت", "۹": "نه" ] for (digit, word) in toWords { t = t.replacingOccurrences(of: digit, with: word) } return t } } private func normalizeText(_ text: String) -> String { var t = text while t.contains(" ") { t = t.replacingOccurrences(of: " ", with: " ") } let punctuationList = [".", "،", "؛", "؟", "!", ":", ")", "»"] for p in punctuationList { t = t.replacingOccurrences(of: " " + p, with: p) } t = t.replacingOccurrences(of: "...", with: "…") for p in ["،", "؛", "؟", "!", "…"] { t = t.replacingOccurrences(of: p, with: p + " ") } // Only add space after "." when it is NOT between two digits (preserve 4.3, 3.8, IP addresses) let dotNotDecimal = try! NSRegularExpression(pattern: #"(\.)(?![0-9۰-۹\s\.])"#, options: []) t = dotNotDecimal.stringByReplacingMatches(in: t, options: [], range: NSRange(location: 0, length: t.utf16.count), withTemplate: ". ") t = t.replacingOccurrences(of: "…", with: "...") let speechFixes: [(String, String)] = [ (" ماا ", " ما "), (" اینپیگیری ", " این پیگیری "), (" می‌کنکنیم ", " می‌کنیم "), (" سر سمانش ", " سر و سامانش "), (" دست و پا گیرلی ", " دست و پا گیر ولی ") ] for (bad, good) in speechFixes { t = t.replacingOccurrences(of: bad, with: good) } t = t.replacingOccurrences(of: "( ", with: "(") t = t.replacingOccurrences(of: "« ", with: "«") while t.contains(" ") { t = t.replacingOccurrences(of: " ", with: " ") } return t.trimmingCharacters(in: .whitespacesAndNewlines) } public func stripTrailingEndTrigger(_ text: String) -> String { var t = text.trimmingCharacters(in: .whitespacesAndNewlines) let range1 = NSRange(location: 0, length: t.utf16.count) t = TextProcessor.trailingEndRegex.stringByReplacingMatches(in: t, options: [], range: range1, withTemplate: "") let range2 = NSRange(location: 0, length: t.utf16.count) t = TextProcessor.singleEndRegex.stringByReplacingMatches(in: t, options: [], range: range2, withTemplate: "") return t.trimmingCharacters(in: .whitespacesAndNewlines) } }