import { exec } from 'child_process'
import { access, constants } from 'fs/promises'
import { promisify } from 'util'
import type { SubtitleResult, LanguageCode, YouTubeVideoInfo } from './types'
import { detectVideoLanguage } from './language-detector'
import { cleanTranscript } from './transcript-cleaner'
import { parseVTTFile } from './vtt-parser'
const execAsync = promisify(exec)
const expandLanguageCodes = (languages: LanguageCode[]): string[] =>
languages.flatMap(lang => {
const expansions: Record<LanguageCode, string[]> = {
ko: ['ko', 'kr'],
ja: ['ja', 'jp'],
zh: ['zh', 'zh-CN', 'zh-TW', 'zh-Hans', 'zh-Hant'],
en: ['en', 'en-US', 'en-GB'],
th: ['th'],
ar: ['ar'],
hi: ['hi'],
ru: ['ru'],
el: ['el'],
}
return expansions[lang] || [lang]
})
const buildSubtitleCommand = (url: string, languages: string[], tempId: string): string => {
const args = [
'yt-dlp',
'--write-subs',
'--write-auto-subs',
'--sub-format vtt',
'--skip-download',
'--no-warnings',
`--sub-langs "${languages.join(',')}"`,
'-o',
`"${tempId}.%(ext)s"`,
`"${url}"`,
]
return args.join(' ')
}
const generateSubtitleFilenames = (tempId: string, lang: string): string[] => [
`${tempId}.${lang}.vtt`,
`${tempId}.${lang}-auto.vtt`,
`${tempId}.${lang}.auto.vtt`,
]
const findSubtitleFile = async (
tempId: string,
languages: string[],
): Promise<null | { file: string; language: string }> => {
for (const lang of languages) {
const possibleFiles = generateSubtitleFilenames(tempId, lang)
for (const file of possibleFiles) {
try {
await access(file, constants.F_OK)
console.log(`[subtitle-extractor] Found subtitle file: ${file}`)
return { file, language: lang }
} catch {
// File doesn't exist, continue to next
}
}
}
return null
}
const findAnySubtitleFile = async (tempId: string): Promise<null | { file: string; language: string }> => {
try {
const { stdout } = await execAsync(`ls ${tempId}*.vtt 2>/dev/null || true`)
const files = stdout
.trim()
.split('\n')
.filter(file => file.length > 0)
if (files.length > 0) {
const [firstFile] = files
const languageMatch = firstFile.match(/\.([a-z-]+)\.vtt$/)
const language = languageMatch?.[1] || 'unknown'
console.log(`[subtitle-extractor] Found any subtitle file: ${firstFile} (${language})`)
return { file: firstFile, language }
}
} catch (error) {
console.log(`[subtitle-extractor] No subtitle files found:`, error)
}
return null
}
const cleanupSubtitleFiles = async (tempId: string): Promise<void> => {
try {
await execAsync(`rm -f ${tempId}*.vtt ${tempId}*.info.json 2>/dev/null || true`)
} catch {
// Ignore cleanup errors
}
}
export const extractSubtitles = async (
url: string,
videoMetadata?: Partial<YouTubeVideoInfo>,
): Promise<SubtitleResult> => {
const tempId = `temp_subs_${Date.now()}`
try {
const languagePriority = videoMetadata
? detectVideoLanguage(videoMetadata)
: (['en', 'ko', 'ja', 'zh', 'es', 'fr'] as LanguageCode[])
console.log(`🔍 Language priority for subtitles: ${languagePriority.join(', ')}`)
const expandedLanguages = expandLanguageCodes(languagePriority)
const command = buildSubtitleCommand(url, expandedLanguages, tempId)
await execAsync(command)
const foundFile = (await findSubtitleFile(tempId, expandedLanguages)) || (await findAnySubtitleFile(tempId))
if (!foundFile) {
throw new Error('No subtitle files found')
}
console.log(`Found subtitle file: ${foundFile.file} (language: ${foundFile.language})`)
const rawTranscript = await parseVTTFile(foundFile.file)
const cleanedTranscript = cleanTranscript(rawTranscript)
await cleanupSubtitleFiles(tempId)
console.log(`📝 Subtitle extraction successful (${foundFile.language})`)
console.log(`📊 Original length: ${rawTranscript.length} chars`)
console.log(`📊 Cleaned length: ${cleanedTranscript.length} chars`)
return {
transcript: cleanedTranscript,
source: 'subtitles',
language: foundFile.language,
}
} catch (error) {
await cleanupSubtitleFiles(tempId)
const errorMessage = error instanceof Error ? error.message : 'Unknown error'
throw new Error(`Failed to extract subtitles: ${errorMessage}`)
}
}
