| 1 | import { Logger } from '@nestjs/common' |
| 2 | import { |
| 3 | AITranslationSkillParams, |
| 4 | LanguageCode, |
| 5 | SkillType, |
| 6 | SubtitleRecognitionType, |
| 7 | TranslationType, |
| 8 | } from './volcengine.interface' |
| 9 | |
| 10 | const logger = new Logger('PromptParser') |
| 11 | |
| 12 | /** |
| 13 | * 语言映射表（中文关键词 -> LanguageCode） |
| 14 | */ |
| 15 | const LANGUAGE_MAP: Record<string, LanguageCode> = { |
| 16 | // 中文 |
| 17 | 中文: LanguageCode.Zh, |
| 18 | 汉语: LanguageCode.Zh, |
| 19 | 中: LanguageCode.Zh, |
| 20 | chinese: LanguageCode.Zh, |
| 21 | zh: LanguageCode.Zh, |
| 22 | |
| 23 | // 英文 |
| 24 | 英文: LanguageCode.En, |
| 25 | 英语: LanguageCode.En, |
| 26 | 英: LanguageCode.En, |
| 27 | english: LanguageCode.En, |
| 28 | en: LanguageCode.En, |
| 29 | |
| 30 | // 日语 |
| 31 | 日文: LanguageCode.Ja, |
| 32 | 日语: LanguageCode.Ja, |
| 33 | 日: LanguageCode.Ja, |
| 34 | japanese: LanguageCode.Ja, |
| 35 | ja: LanguageCode.Ja, |
| 36 | |
| 37 | // 韩语 |
| 38 | 韩文: LanguageCode.Ko, |
| 39 | 韩语: LanguageCode.Ko, |
| 40 | 韩: LanguageCode.Ko, |
| 41 | 朝鲜语: LanguageCode.Ko, |
| 42 | korean: LanguageCode.Ko, |
| 43 | ko: LanguageCode.Ko, |
| 44 | |
| 45 | // 德语 |
| 46 | 德文: LanguageCode.De, |
| 47 | 德语: LanguageCode.De, |
| 48 | 德: LanguageCode.De, |
| 49 | german: LanguageCode.De, |
| 50 | de: LanguageCode.De, |
| 51 | |
| 52 | // 法语 |
| 53 | 法文: LanguageCode.Fr, |
| 54 | 法语: LanguageCode.Fr, |
| 55 | 法: LanguageCode.Fr, |
| 56 | french: LanguageCode.Fr, |
| 57 | fr: LanguageCode.Fr, |
| 58 | |
| 59 | // 俄语 |
| 60 | 俄文: LanguageCode.Ru, |
| 61 | 俄语: LanguageCode.Ru, |
| 62 | 俄: LanguageCode.Ru, |
| 63 | russian: LanguageCode.Ru, |
| 64 | ru: LanguageCode.Ru, |
| 65 | |
| 66 | // 西班牙语 |
| 67 | 西班牙文: LanguageCode.Es, |
| 68 | 西班牙语: LanguageCode.Es, |
| 69 | 西语: LanguageCode.Es, |
| 70 | spanish: LanguageCode.Es, |
| 71 | es: LanguageCode.Es, |
| 72 | |
| 73 | // 葡萄牙语 |
| 74 | 葡萄牙文: LanguageCode.Pt, |
| 75 | 葡萄牙语: LanguageCode.Pt, |
| 76 | 葡语: LanguageCode.Pt, |
| 77 | portuguese: LanguageCode.Pt, |
| 78 | pt: LanguageCode.Pt, |
| 79 | |
| 80 | // 意大利语 |
| 81 | 意大利文: LanguageCode.It, |
| 82 | 意大利语: LanguageCode.It, |
| 83 | 意语: LanguageCode.It, |
| 84 | italian: LanguageCode.It, |
| 85 | it: LanguageCode.It, |
| 86 | |
| 87 | // 印尼语 |
| 88 | 印尼文: LanguageCode.Id, |
| 89 | 印尼语: LanguageCode.Id, |
| 90 | 印度尼西亚语: LanguageCode.Id, |
| 91 | indonesian: LanguageCode.Id, |
| 92 | id: LanguageCode.Id, |
| 93 | |
| 94 | // 越南语 |
| 95 | 越南文: LanguageCode.Vi, |
| 96 | 越南语: LanguageCode.Vi, |
| 97 | 越语: LanguageCode.Vi, |
| 98 | vietnamese: LanguageCode.Vi, |
| 99 | vi: LanguageCode.Vi, |
| 100 | |
| 101 | // 泰语 |
| 102 | 泰文: LanguageCode.Th, |
| 103 | 泰语: LanguageCode.Th, |
| 104 | 泰: LanguageCode.Th, |
| 105 | thai: LanguageCode.Th, |
| 106 | th: LanguageCode.Th, |
| 107 | |
| 108 | // 阿拉伯语 |
| 109 | 阿拉伯文: LanguageCode.Ar, |
| 110 | 阿拉伯语: LanguageCode.Ar, |
| 111 | 阿语: LanguageCode.Ar, |
| 112 | arabic: LanguageCode.Ar, |
| 113 | ar: LanguageCode.Ar, |
| 114 | |
| 115 | // 土耳其语 |
| 116 | 土耳其文: LanguageCode.Tr, |
| 117 | 土耳其语: LanguageCode.Tr, |
| 118 | 土语: LanguageCode.Tr, |
| 119 | turkish: LanguageCode.Tr, |
| 120 | tr: LanguageCode.Tr, |
| 121 | } |
| 122 | |
| 123 | const DEFAULT_SOURCE: LanguageCode = LanguageCode.Zh |
| 124 | const DEFAULT_TARGET: LanguageCode = LanguageCode.En |
| 125 | // Note: Only FacialTranslation is supported, no keyword mapping needed |
| 126 | |
| 127 | /** |
| 128 | * from prompt extract source and target languages |
| 129 | */ |
| 130 | /** |
| 131 | * 清洗并标准化语言关键词 |
| 132 | */ |
| 133 | function normalizeLang(raw?: string): LanguageCode | undefined { |
| 134 | if (!raw) |
| 135 | return |
| 136 | const key = raw |
| 137 | .toLowerCase() |
| 138 | .replace(/[^\w\u4E00-\u9FA5]/g, '') // 去掉标点，只保留字母/数字/中文 |
| 139 | return LANGUAGE_MAP[key] |
| 140 | } |
| 141 | |
| 142 | /** |
| 143 | * 从 prompt 中提取源语言和目标语言 |
| 144 | */ |
| 145 | function extractLanguage(prompt: string): { source: LanguageCode, target: LanguageCode } { |
| 146 | let source: LanguageCode | undefined |
| 147 | let target: LanguageCode | undefined |
| 148 | |
| 149 | // 常见中英文表达 |
| 150 | const patterns: RegExp[] = [ |
| 151 | // 从中文翻译到英文 / 由A翻译成B |
| 152 | /(?:从|由)\s*([^\s到成为至:：,，。]+)\s*(?:翻译)?[到成为至]\s*([^\s:：,，。]+)/, |
| 153 | |
| 154 | // 中文翻译成英文 / 中文译成英文 |
| 155 | /(\S+)\s*(?:翻译|译)[成为到]\s*([^\s:：,，。]+)/, |
| 156 | |
| 157 | // translate from chinese to english |
| 158 | /translate\s+(?:from\s+)?(\w+)\s+(?:to|into)\s+(\w+)/i, |
| 159 | ] |
| 160 | |
| 161 | for (const p of patterns) { |
| 162 | const m = prompt.match(p) |
| 163 | if (m) { |
| 164 | source = normalizeLang(m[1]) |
| 165 | target = normalizeLang(m[2]) |
| 166 | if (source || target) |
| 167 | break |
| 168 | } |
| 169 | } |
| 170 | |
| 171 | // 只指定目标语言：翻译为英文 / translate to english |
| 172 | if (!target) { |
| 173 | const targetPatterns: RegExp[] = [ |
| 174 | /(?:翻译|译)[为成到]\s*([^\s:：,，。]+)/, |
| 175 | /translate\s+(?:to|into)\s+(\w+)/i, |
| 176 | ] |
| 177 | |
| 178 | for (const p of targetPatterns) { |
| 179 | const m = prompt.match(p) |
| 180 | if (m) { |
| 181 | target = normalizeLang(m[1]) |
| 182 | if (target) |
| 183 | break |
| 184 | } |
| 185 | } |
| 186 | } |
| 187 | |
| 188 | // 🔹 MCP 提示词兜底：to target language |
| 189 | if (!target && /target language/i.test(prompt)) { |
| 190 | target = DEFAULT_TARGET |
| 191 | } |
| 192 | |
| 193 | // ✅ 默认：中文 → 英文 |
| 194 | source = source ?? DEFAULT_SOURCE |
| 195 | target = target ?? DEFAULT_TARGET |
| 196 | |
| 197 | logger.debug(`Extracted language => source: ${source}, target: ${target}`) |
| 198 | return { source, target } |
| 199 | } |
| 200 | |
| 201 | /** |
| 202 | * Extract translation type from prompt |
| 203 | */ |
| 204 | function extractTranslationType(_prompt: string): TranslationType[] { |
| 205 | logger.debug('Using FacialTranslation (only supported type)') |
| 206 | return [TranslationType.FacialTranslation, TranslationType.SubtitleTranslation, TranslationType.VoiceTranslation] |
| 207 | } |
| 208 | |
| 209 | /** |
| 210 | * Extract subtitle recognition type from prompt |
| 211 | */ |
| 212 | function extractSubtitleRecognitionType(prompt: string): SubtitleRecognitionType { |
| 213 | const lowerPrompt = prompt.toLowerCase() |
| 214 | |
| 215 | if (lowerPrompt.includes('ocr') || lowerPrompt.includes('image')) { |
| 216 | return SubtitleRecognitionType.OCR |
| 217 | } |
| 218 | |
| 219 | if (lowerPrompt.includes('asr') || lowerPrompt.includes('audio') || lowerPrompt.includes('speech')) { |
| 220 | return SubtitleRecognitionType.ASR |
| 221 | } |
| 222 | |
| 223 | if (lowerPrompt.includes('vision') || lowerPrompt.includes('ai')) { |
| 224 | return SubtitleRecognitionType.Vision |
| 225 | } |
| 226 | |
| 227 | // 如果提到硬字幕相关关键词，使用 OCR |
| 228 | if (lowerPrompt.includes('硬字幕') || lowerPrompt.includes('hard subtitle') || lowerPrompt.includes('hardcoded')) { |
| 229 | return SubtitleRecognitionType.OCR |
| 230 | } |
| 231 | |
| 232 | // Default to ASR - 大多数视频翻译场景是翻译语音内容 |
| 233 | return SubtitleRecognitionType.ASR |
| 234 | } |
| 235 | |
| 236 | /** |
| 237 | * Check if hard subtitle and source erasure are needed |
| 238 | * (Reserved for future prompt-based override) |
| 239 | */ |
| 240 | function _extractSubtitleConfig(prompt: string) { |
| 241 | const lowerPrompt = prompt.toLowerCase() |
| 242 | |
| 243 | const needHardSubtitle = lowerPrompt.includes('hard subtitle') |
| 244 | || lowerPrompt.includes('burn') |
| 245 | || lowerPrompt.includes('embed') |
| 246 | |
| 247 | const needEraseSource = lowerPrompt.includes('erase') |
| 248 | || lowerPrompt.includes('remove') |
| 249 | || needHardSubtitle |
| 250 | |
| 251 | return { |
| 252 | needHardSubtitle, |
| 253 | needEraseSource, |
| 254 | } |
| 255 | } |
| 256 | |
| 257 | function isTranslationPrompt(prompt: string): boolean { |
| 258 | const text = prompt.toLowerCase() |
| 259 | |
| 260 | const patterns: RegExp[] = [ |
| 261 | // 中文翻译意图 |
| 262 | /翻译/, |
| 263 | /译为|译成/, |
| 264 | /将.*翻译/, |
| 265 | /字幕.*翻译/, |
| 266 | /翻译.*字幕/, |
| 267 | |
| 268 | // 英文翻译意图 |
| 269 | /translate/, |
| 270 | /translation/, |
| 271 | /subtitles?/, |
| 272 | /voice/, |
| 273 | /lip/, |
| 274 | /facial/, |
| 275 | /audio/, |
| 276 | ] |
| 277 | |
| 278 | return patterns.some(p => p.test(text)) |
| 279 | } |
| 280 | |
| 281 | /** |
| 282 | * Parse translation prompt |
| 283 | */ |
| 284 | export function parseTranslationPrompt(prompt: string): AITranslationSkillParams | null { |
| 285 | logger.debug(`Parsing translation prompt: ${prompt}`) |
| 286 | |
| 287 | if (!isTranslationPrompt(prompt)) { |
| 288 | logger.debug('Prompt does not contain translation intent') |
| 289 | return null |
| 290 | } |
| 291 | |
| 292 | const { source, target } = extractLanguage(prompt) |
| 293 | |
| 294 | if (!target) { |
| 295 | logger.warn('Unable to extract target language from prompt') |
| 296 | return null |
| 297 | } |
| 298 | |
| 299 | const translationTypes = extractTranslationType(prompt) |
| 300 | const recognitionType = extractSubtitleRecognitionType(prompt) |
| 301 | |
| 302 | // Build params - Facial translation config |
| 303 | const params: AITranslationSkillParams = { |
| 304 | TranslationConfig: { |
| 305 | SourceLanguage: source || LanguageCode.Zh, // Default source: Chinese |
| 306 | TargetLanguage: target, |
| 307 | TranslationTypeList: translationTypes, // Always [FacialTranslation] |
| 308 | }, |
| 309 | OperatorConfig: { |
| 310 | SubtitleRecognitionConfig: { |
| 311 | RecognitionType: recognitionType, |
| 312 | }, |
| 313 | }, |
| 314 | } |
| 315 | |
| 316 | // Facial translation defaults: hard subtitle + erase source |
| 317 | params.SubtitleConfig = { |
| 318 | IsHardSubtitle: true, |
| 319 | IsEraseSource: true, |
| 320 | FontSize: 24, |
| 321 | MarginL: 0.1, |
| 322 | MarginR: 0.1, |
| 323 | MarginV: 0.05, |
| 324 | ShowLines: 2, |
| 325 | } |
| 326 | logger.debug('Applied facial translation defaults: hard subtitle + erase source') |
| 327 | |
| 328 | return params |
| 329 | } |
| 330 | |
| 331 | /** |
| 332 | * Parse VCreative (AI editing) prompt |
| 333 | * Currently disabled, reserved for future use |
| 334 | */ |
| 335 | function _parseVCreativePrompt(prompt: string): { Text: string } | null { |
| 336 | const lowerPrompt = prompt.toLowerCase() |
| 337 | |
| 338 | const mergeKeywords = [ |
| 339 | 'merge', |
| 340 | 'combine', |
| 341 | 'mix', |
| 342 | 'audio', |
| 343 | 'video', |
| 344 | 'sound', |
| 345 | ] |
| 346 | |
| 347 | const isMerge = mergeKeywords.some(keyword => lowerPrompt.includes(keyword)) |
| 348 | |
| 349 | if (isMerge) { |
| 350 | logger.debug('Detected merge-related prompt') |
| 351 | |
| 352 | let optimizedText = '' |
| 353 | |
| 354 | if (lowerPrompt.includes('audio') || lowerPrompt.includes('sound')) { |
| 355 | optimizedText = 'Merge video visuals with audio from second input' |
| 356 | } |
| 357 | else if (lowerPrompt.includes('concat') || lowerPrompt.includes('join')) { |
| 358 | optimizedText = prompt |
| 359 | } |
| 360 | else { |
| 361 | optimizedText = prompt |
| 362 | } |
| 363 | |
| 364 | logger.debug(`VCreative prompt optimized: ${prompt} -> ${optimizedText}`) |
| 365 | |
| 366 | return { |
| 367 | Text: optimizedText, |
| 368 | } |
| 369 | } |
| 370 | |
| 371 | const vCreativeKeywords = [ |
| 372 | 'concat', |
| 373 | 'trim', |
| 374 | 'cut', |
| 375 | 'edit', |
| 376 | 'image', |
| 377 | 'generate video', |
| 378 | ] |
| 379 | |
| 380 | const isVCreative = vCreativeKeywords.some(keyword => lowerPrompt.includes(keyword)) |
| 381 | |
| 382 | if (isVCreative) { |
| 383 | logger.debug('Detected VCreative prompt') |
| 384 | return { |
| 385 | Text: prompt, |
| 386 | } |
| 387 | } |
| 388 | |
| 389 | return null |
| 390 | } |
| 391 | |
| 392 | /** |
| 393 | * Check if prompt is a subtitle erase task (no translation intent) |
| 394 | */ |
| 395 | function isEraseOnlyTask(prompt: string): boolean { |
| 396 | const lowerPrompt = prompt.toLowerCase() |
| 397 | |
| 398 | const eraseKeywords = ['remove', 'erase', 'delete', 'clear'] |
| 399 | const hasEraseKeyword = eraseKeywords.some(keyword => lowerPrompt.includes(keyword)) |
| 400 | |
| 401 | if (!hasEraseKeyword) { |
| 402 | return false |
| 403 | } |
| 404 | |
| 405 | // Subtitle-related keywords |
| 406 | const subtitleKeywords = ['subtitle', 'caption', 'watermark'] |
| 407 | const hasSubtitleKeyword = subtitleKeywords.some(keyword => lowerPrompt.includes(keyword)) |
| 408 | |
| 409 | // Exclude translation intent |
| 410 | const translationIntentKeywords = ['translate', 'translation'] |
| 411 | const hasTranslationIntent = translationIntentKeywords.some(keyword => lowerPrompt.includes(keyword)) |
| 412 | |
| 413 | // Erase task: has erase + subtitle keywords, but no translation intent |
| 414 | return hasSubtitleKeyword && !hasTranslationIntent |
| 415 | } |
| 416 | |
| 417 | /** |
| 418 | * Check if prompt is a highlight extraction task |
| 419 | */ |
| 420 | function isHighlightTask(prompt: string): boolean { |
| 421 | const lowerPrompt = prompt.toLowerCase() |
| 422 | |
| 423 | const highlightKeywords = [ |
| 424 | 'highlight', |
| 425 | 'exciting', |
| 426 | 'best moments', |
| 427 | 'compilation', |
| 428 | 'montage', |
| 429 | ] |
| 430 | |
| 431 | return highlightKeywords.some(keyword => lowerPrompt.includes(keyword)) |
| 432 | } |
| 433 | |
| 434 | /** |
| 435 | * Check if prompt is a video understanding task |
| 436 | */ |
| 437 | function isVisionTask(prompt: string): boolean { |
| 438 | const lowerPrompt = prompt.toLowerCase() |
| 439 | |
| 440 | const visionKeywords = [ |
| 441 | 'understand', |
| 442 | 'summarize', |
| 443 | 'summary', |
| 444 | 'analyze', |
| 445 | 'analysis', |
| 446 | 'describe', |
| 447 | 'what is', |
| 448 | 'about', |
| 449 | 'content', |
| 450 | 'keywords', |
| 451 | 'tags', |
| 452 | ] |
| 453 | |
| 454 | return visionKeywords.some(keyword => lowerPrompt.includes(keyword)) |
| 455 | } |
| 456 | |
| 457 | /** |
| 458 | * Check if prompt is a VCreative (AI editing) task |
| 459 | */ |
| 460 | function isVCreativeTask(prompt: string): boolean { |
| 461 | const lowerPrompt = prompt.toLowerCase() |
| 462 | |
| 463 | const vCreativeKeywords = [ |
| 464 | 'concat', |
| 465 | 'merge', |
| 466 | 'combine', |
| 467 | 'edit', |
| 468 | 'trim', |
| 469 | 'cut', |
| 470 | 'splice', |
| 471 | 'join', |
| 472 | 'add music', |
| 473 | 'background music', |
| 474 | 'picture-in-picture', |
| 475 | 'pip', |
| 476 | ] |
| 477 | |
| 478 | return vCreativeKeywords.some(keyword => lowerPrompt.includes(keyword)) |
| 479 | } |
| 480 | |
| 481 | /** |
| 482 | * Parse Aideo task prompt to extract skill type and parameters |
| 483 | * |
| 484 | * Priority order (high to low): |
| 485 | * 1. AITranslation - Translation tasks |
| 486 | * 2. Erase - Subtitle/watermark removal |
| 487 | * 3. Highlight - Highlight extraction |
| 488 | * 4. Vision - Video understanding |
| 489 | * 5. VCreative - AI editing (fallback) |
| 490 | */ |
| 491 | export function parseAideoPrompt(prompt: string): { |
| 492 | skillType: SkillType |
| 493 | skillParams: Record<string, unknown> |
| 494 | } | null { |
| 495 | logger.debug(`Parsing Aideo prompt: ${prompt}`) |
| 496 | |
| 497 | // 1. Check translation task first |
| 498 | const translationParams = parseTranslationPrompt(prompt) |
| 499 | if (translationParams) { |
| 500 | logger.debug('Detected AITranslation task') |
| 501 | return { |
| 502 | skillType: SkillType.AITranslation, |
| 503 | skillParams: translationParams as unknown as Record<string, unknown>, |
| 504 | } |
| 505 | } |
| 506 | |
| 507 | // 2. Check subtitle erase task |
| 508 | if (isEraseOnlyTask(prompt)) { |
| 509 | logger.debug('Detected Erase task') |
| 510 | return { |
| 511 | skillType: SkillType.Erase, |
| 512 | skillParams: { |
| 513 | Mode: 'Auto', |
| 514 | Auto: { Type: 'Subtitle' }, |
| 515 | NewVid: true, |
| 516 | }, |
| 517 | } |
| 518 | } |
| 519 | |
| 520 | // 3. Check highlight task |
| 521 | if (isHighlightTask(prompt)) { |
| 522 | logger.debug('Detected Highlight task') |
| 523 | return { |
| 524 | skillType: SkillType.Highlight, |
| 525 | skillParams: { |
| 526 | Text: prompt, |
| 527 | }, |
| 528 | } |
| 529 | } |
| 530 | |
| 531 | // 4. Check vision task |
| 532 | if (isVisionTask(prompt)) { |
| 533 | logger.debug('Detected Vision task') |
| 534 | return { |
| 535 | skillType: SkillType.Vision, |
| 536 | skillParams: { |
| 537 | Text: prompt, |
| 538 | }, |
| 539 | } |
| 540 | } |
| 541 | |
| 542 | // 5. Check VCreative task (fallback) |
| 543 | if (isVCreativeTask(prompt)) { |
| 544 | logger.debug('Detected VCreative task') |
| 545 | return { |
| 546 | skillType: SkillType.VCreative, |
| 547 | skillParams: { |
| 548 | Text: prompt, |
| 549 | }, |
| 550 | } |
| 551 | } |
| 552 | |
| 553 | // No match, let server decide |
| 554 | logger.debug('Unable to detect skill type, server will decide') |
| 555 | return null |
| 556 | } |
| 557 |