|
| 1 | +// |
| 2 | +// Copyright 2025 Readium Foundation. All rights reserved. |
| 3 | +// Use of this source code is governed by the BSD-style license |
| 4 | +// available in the top-level LICENSE file of the project. |
| 5 | +// |
| 6 | + |
| 7 | +import AVFoundation |
| 8 | +import Foundation |
| 9 | +import ReadiumShared |
| 10 | + |
| 11 | +/// Represents a voice provided by the TTS engine which can speak an utterance. |
| 12 | +public struct TTSVoice: Hashable { |
| 13 | + public enum Gender: Hashable { |
| 14 | + case female, male, unspecified |
| 15 | + } |
| 16 | + |
| 17 | + public enum Quality: Hashable { |
| 18 | + case lower, low, medium, high, higher |
| 19 | + } |
| 20 | + |
| 21 | + /// Unique and stable identifier for this voice. Can be used to store and retrieve the voice from the user |
| 22 | + /// preferences. |
| 23 | + public let identifier: String |
| 24 | + |
| 25 | + /// Human-friendly name for this voice. |
| 26 | + public let name: String |
| 27 | + |
| 28 | + /// Language (and region) this voice belongs to. |
| 29 | + public let language: Language |
| 30 | + |
| 31 | + /// Voice gender. |
| 32 | + public let gender: Gender |
| 33 | + |
| 34 | + /// Voice quality. |
| 35 | + public let quality: Quality? |
| 36 | + |
| 37 | + public init( |
| 38 | + identifier: String, |
| 39 | + language: Language, |
| 40 | + name: String, |
| 41 | + gender: Gender, |
| 42 | + quality: Quality? |
| 43 | + ) { |
| 44 | + self.identifier = identifier |
| 45 | + self.language = language |
| 46 | + self.name = name |
| 47 | + self.gender = gender |
| 48 | + self.quality = quality |
| 49 | + } |
| 50 | +} |
| 51 | + |
| 52 | +public extension [TTSVoice] { |
| 53 | + /// Filter voices by language. |
| 54 | + /// |
| 55 | + /// If the input language includes a region (e.g., "en-US"), only voices |
| 56 | + /// with that exact language and region combination will be returned. |
| 57 | + /// If the input language has no region (e.g., "en"), all voices matching |
| 58 | + /// the base language will be returned, regardless of their region. |
| 59 | + func filterByLanguage(_ language: Language) -> [TTSVoice] { |
| 60 | + if language.region != nil { |
| 61 | + // Exact match: language code and region must both match |
| 62 | + return filter { $0.language == language } |
| 63 | + } else { |
| 64 | + // Base language match: filter by language code only, ignoring region |
| 65 | + return filter { $0.language.removingRegion() == language } |
| 66 | + } |
| 67 | + } |
| 68 | + |
| 69 | + /// Sort the voices according to the following specification: |
| 70 | + /// 1. Order by region: |
| 71 | + /// - Use the regions of the devices at the top of the list. |
| 72 | + /// - If missing, use a default region per language. |
| 73 | + /// - Then order the remaining regions alphabetically. |
| 74 | + /// 2. Order by voice quality from highest to lowest. |
| 75 | + /// 3. Order by voice gender: female > male > unspecified. |
| 76 | + /// 4. Order by voice name alphabetically. |
| 77 | + func sorted( |
| 78 | + preferredRegions: [Language.Region]? = nil, |
| 79 | + displayLocale: Locale = .current |
| 80 | + ) -> [TTSVoice] { |
| 81 | + let languagesAndRegions: [(language: Language, region: Language.Region?)] = |
| 82 | + map { ($0.language.removingRegion(), $0.language.region) } |
| 83 | + |
| 84 | + let regionsByLanguage: [Language: Set<Language.Region>] = |
| 85 | + Dictionary(grouping: languagesAndRegions, by: \.language) |
| 86 | + .mapValues { value in |
| 87 | + Set(value.compactMap(\.region)) |
| 88 | + } |
| 89 | + |
| 90 | + let preferredRegions = preferredRegions ?? Locale.preferredRegions |
| 91 | + |
| 92 | + let regionPrioritiesByLanguage: [Language: [Language.Region: Int]] = |
| 93 | + Dictionary(uniqueKeysWithValues: regionsByLanguage.map { language, regions in |
| 94 | + var ordered: [Language.Region] = [] |
| 95 | + |
| 96 | + // 1. Start with device-preferred regions. |
| 97 | + ordered.append( |
| 98 | + contentsOf: preferredRegions.filter { regions.contains($0) } |
| 99 | + ) |
| 100 | + |
| 101 | + // 2. Default region for the language, if any. |
| 102 | + if let defaultRegion = defaultRegionByLanguage[language.code] { |
| 103 | + ordered.append(defaultRegion) |
| 104 | + } |
| 105 | + |
| 106 | + // 3. Add remaining regions ordered by localized name. |
| 107 | + ordered.append(contentsOf: regions.sorted { |
| 108 | + ($0.localizedName(in: displayLocale) ?? $0.code) < ($1.localizedName(in: displayLocale) ?? $1.code) |
| 109 | + } |
| 110 | + ) |
| 111 | + |
| 112 | + ordered = ordered.removingDuplicates() |
| 113 | + |
| 114 | + // Assign priorities: lower Int = higher priority |
| 115 | + let priorities = Dictionary(uniqueKeysWithValues: |
| 116 | + ordered.enumerated().map { idx, region in (region, idx) } |
| 117 | + ) |
| 118 | + |
| 119 | + return (language, priorities) |
| 120 | + }) |
| 121 | + |
| 122 | + func sortKey(for voice: TTSVoice) -> ( |
| 123 | + language: String, |
| 124 | + region: Int, |
| 125 | + quality: Int, |
| 126 | + gender: Int, |
| 127 | + name: String |
| 128 | + ) { |
| 129 | + let language = voice.language.removingRegion() |
| 130 | + |
| 131 | + let regionPriority: Int = |
| 132 | + if |
| 133 | + let region = voice.language.region, |
| 134 | + let regionPriorities = regionPrioritiesByLanguage[language] |
| 135 | + { |
| 136 | + regionPriorities[region] ?? .max |
| 137 | + } else { |
| 138 | + .max |
| 139 | + } |
| 140 | + |
| 141 | + return ( |
| 142 | + language: language.localizedLanguage(in: displayLocale) ?? voice.language.code.bcp47, |
| 143 | + region: regionPriority, |
| 144 | + quality: voice.quality.flatMap { qualityPriorities[$0] } ?? .max, |
| 145 | + gender: genderPriorities[voice.gender] ?? .max, |
| 146 | + name: voice.name |
| 147 | + ) |
| 148 | + } |
| 149 | + |
| 150 | + let voicesWithKeys = map { voice in |
| 151 | + (voice: voice, key: sortKey(for: voice)) |
| 152 | + } |
| 153 | + |
| 154 | + return voicesWithKeys.sorted { a, b in |
| 155 | + let ka = a.key |
| 156 | + let kb = b.key |
| 157 | + |
| 158 | + if ka.language != kb.language { |
| 159 | + return ka.language < kb.language |
| 160 | + } |
| 161 | + if ka.region != kb.region { |
| 162 | + return ka.region < kb.region |
| 163 | + } |
| 164 | + if ka.quality != kb.quality { |
| 165 | + return ka.quality < kb.quality |
| 166 | + } |
| 167 | + if ka.gender != kb.gender { |
| 168 | + return ka.gender < kb.gender |
| 169 | + } |
| 170 | + |
| 171 | + return ka.name.localizedCaseInsensitiveCompare(kb.name) == .orderedAscending |
| 172 | + }.map(\.voice) |
| 173 | + } |
| 174 | +} |
| 175 | + |
| 176 | +private extension Locale { |
| 177 | + static var preferredRegions: [ReadiumShared.Language.Region] { |
| 178 | + preferredLanguages |
| 179 | + .compactMap { ReadiumShared.Language(code: .bcp47($0)).region } |
| 180 | + } |
| 181 | +} |
| 182 | + |
| 183 | +// Default region per base language. |
| 184 | +// Source: https://github.com/HadrienGardeur/web-speech-recommended-voices |
| 185 | +private let defaultRegionByLanguage: [Language.Code: Language.Region] = [ |
| 186 | + .bcp47("ar"): "SA", |
| 187 | + .bcp47("bg"): "BG", |
| 188 | + .bcp47("bho"): "IN", |
| 189 | + .bcp47("bn"): "IN", |
| 190 | + .bcp47("ca"): "ES", |
| 191 | + .bcp47("cmn"): "CN", |
| 192 | + .bcp47("cs"): "CZ", |
| 193 | + .bcp47("da"): "DK", |
| 194 | + .bcp47("de"): "DE", |
| 195 | + .bcp47("el"): "GR", |
| 196 | + .bcp47("en"): "US", |
| 197 | + .bcp47("es"): "ES", |
| 198 | + .bcp47("eu"): "ES", |
| 199 | + .bcp47("fa"): "IR", |
| 200 | + .bcp47("fi"): "FI", |
| 201 | + .bcp47("fr"): "FR", |
| 202 | + .bcp47("gl"): "ES", |
| 203 | + .bcp47("he"): "IL", |
| 204 | + .bcp47("hi"): "IN", |
| 205 | + .bcp47("hr"): "HR", |
| 206 | + .bcp47("hu"): "HU", |
| 207 | + .bcp47("id"): "ID", |
| 208 | + .bcp47("it"): "IT", |
| 209 | + .bcp47("ja"): "JP", |
| 210 | + .bcp47("kn"): "IN", |
| 211 | + .bcp47("ko"): "KR", |
| 212 | + .bcp47("mr"): "IN", |
| 213 | + .bcp47("ms"): "MY", |
| 214 | + .bcp47("nb"): "NO", |
| 215 | + .bcp47("nl"): "NL", |
| 216 | + .bcp47("pl"): "PL", |
| 217 | + .bcp47("pt"): "BR", |
| 218 | + .bcp47("ro"): "RO", |
| 219 | + .bcp47("ru"): "RU", |
| 220 | + .bcp47("sk"): "SK", |
| 221 | + .bcp47("sl"): "SI", |
| 222 | + .bcp47("sv"): "SE", |
| 223 | + .bcp47("ta"): "IN", |
| 224 | + .bcp47("te"): "IN", |
| 225 | + .bcp47("th"): "TH", |
| 226 | + .bcp47("tr"): "TR", |
| 227 | + .bcp47("uk"): "UA", |
| 228 | + .bcp47("vi"): "VN", |
| 229 | + .bcp47("wuu"): "CN", |
| 230 | + .bcp47("yue"): "HK", |
| 231 | +] |
| 232 | + |
| 233 | +// Quality order priority: higher to lower |
| 234 | +private let qualityPriorities: [TTSVoice.Quality: Int] = [ |
| 235 | + .higher: 0, |
| 236 | + .high: 1, |
| 237 | + .medium: 2, |
| 238 | + .low: 3, |
| 239 | + .lower: 4, |
| 240 | +] |
| 241 | + |
| 242 | +// Gender order priority: female > male > unspecified |
| 243 | +private let genderPriorities: [TTSVoice.Gender: Int] = [ |
| 244 | + .female: 0, |
| 245 | + .male: 1, |
| 246 | + .unspecified: 2, |
| 247 | +] |
0 commit comments