diff --git a/src/app/(app)/epub/[id]/page.tsx b/src/app/(app)/epub/[id]/page.tsx index b08d4d4..3cc6ebd 100644 --- a/src/app/(app)/epub/[id]/page.tsx +++ b/src/app/(app)/epub/[id]/page.tsx @@ -19,6 +19,7 @@ import { RateLimitBanner } from '@/components/auth/RateLimitBanner'; import { useAuthRateLimit } from '@/contexts/AuthRateLimitContext'; import { useFeatureFlag } from '@/contexts/RuntimeConfigContext'; import { useUnmountCleanupRef } from '@/hooks/useUnmountCleanupRef'; +import { useDocumentLanguage } from '@/hooks/useDocumentLanguage'; import { ButtonLink } from '@/components/ui'; import { useEpubDocument } from './useEpubDocument'; @@ -37,7 +38,8 @@ export default function EPUBPage() { regenerateChapter: regenerateEPUBChapter, bookRef, } = epubState; - const { stop } = useTTS(); + const { stop, setDocumentLanguage } = useTTS(); + const { language, updateLanguage } = useDocumentLanguage(routeDocumentId); const { isAtLimit } = useAuthRateLimit(); const [error, setError] = useState(null); const [isLoading, setIsLoading] = useState(true); @@ -106,6 +108,10 @@ export default function EPUBPage() { useUnmountCleanupRef(clearCurrDoc); + useEffect(() => { + setDocumentLanguage(language); + }, [language, setDocumentLanguage]); + // Compute available height = viewport - (header height + tts bar height) useEffect(() => { const compute = () => { @@ -249,6 +255,10 @@ export default function EPUBPage() { epub isOpen={activeSidebar === 'settings'} setIsOpen={(isOpen) => setActiveSidebar((prev) => isOpen ? 'settings' : (prev === 'settings' ? null : prev))} + language={language} + onLanguageChange={(nextLanguage) => { + void updateLanguage(nextLanguage); + }} /> () => { diff --git a/src/app/(app)/html/[id]/page.tsx b/src/app/(app)/html/[id]/page.tsx index 09a0b45..ef0ba89 100644 --- a/src/app/(app)/html/[id]/page.tsx +++ b/src/app/(app)/html/[id]/page.tsx @@ -17,6 +17,7 @@ import { AudiobookExportModal } from '@/components/AudiobookExportModal'; import { useAuthRateLimit } from '@/contexts/AuthRateLimitContext'; import { useFeatureFlag } from '@/contexts/RuntimeConfigContext'; import { useUnmountCleanupRef } from '@/hooks/useUnmountCleanupRef'; +import { useDocumentLanguage } from '@/hooks/useDocumentLanguage'; import { ButtonLink } from '@/components/ui'; import type { TTSAudiobookChapter } from '@/types/tts'; import type { AudiobookGenerationSettings } from '@/types/client'; @@ -26,6 +27,7 @@ export default function HTMLPage() { const canExportAudiobook = useFeatureFlag('enableAudiobookExport'); const { id } = useParams(); const router = useRouter(); + const routeDocumentId = typeof id === 'string' ? id : undefined; const htmlState = useHtmlDocument(); const { setCurrentDocument, @@ -38,7 +40,8 @@ export default function HTMLPage() { createFullAudioBook, regenerateChapter, } = htmlState; - const { stop } = useTTS(); + const { stop, setDocumentLanguage } = useTTS(); + const { language, updateLanguage } = useDocumentLanguage(routeDocumentId); const { isAtLimit } = useAuthRateLimit(); const [error, setError] = useState(null); const [isLoading, setIsLoading] = useState(true); @@ -106,6 +109,10 @@ export default function HTMLPage() { useUnmountCleanupRef(clearCurrDoc); + useEffect(() => { + setDocumentLanguage(language); + }, [language, setDocumentLanguage]); + // Compute available height = viewport - (header height + tts bar height) useEffect(() => { const compute = () => { @@ -240,6 +247,10 @@ export default function HTMLPage() { html isOpen={activeSidebar === 'settings'} setIsOpen={(isOpen) => setActiveSidebar((prev) => isOpen ? 'settings' : (prev === 'settings' ? null : prev))} + language={language} + onLanguageChange={(nextLanguage) => { + void updateLanguage(nextLanguage); + }} /> setActiveSidebar((prev) => isOpen ? 'settings' : (prev === 'settings' ? null : prev))} + language={documentSettings.language ?? 'auto'} + onLanguageChange={(language) => { + void updateDocumentSettings({ + ...documentSettings, + schemaVersion: 1, + language, + }); + }} pdf={{ parseStatus, parsedOverlayEnabled, diff --git a/src/app/(app)/pdf/[id]/usePdfDocument.ts b/src/app/(app)/pdf/[id]/usePdfDocument.ts index 8ec76c4..86cebf9 100644 --- a/src/app/(app)/pdf/[id]/usePdfDocument.ts +++ b/src/app/(app)/pdf/[id]/usePdfDocument.ts @@ -91,6 +91,7 @@ export interface PdfDocumentState { parsedDocument?: ParsedPdfDocument | null; locator?: TTSSegmentLocator | null; useBlockGeometryOnly?: boolean; + language?: string; }, ) => void; clearHighlights: () => void; @@ -134,6 +135,7 @@ export function usePdfDocument(): PdfDocumentState { currDocPages, setCurrDocPages, setIsEPUB, + setDocumentLanguage, registerVisualPageChangeHandler, } = useTTS(); const { @@ -156,6 +158,10 @@ export function usePdfDocument(): PdfDocumentState { const [parseProgress, setParseProgress] = useState(null); const [, setActiveParseOpId] = useState(null); const [documentSettings, setDocumentSettings] = useState(DEFAULT_DOCUMENT_SETTINGS); + useEffect(() => { + setDocumentLanguage(documentSettings.language ?? 'auto'); + lastPreparedPlaybackPageRef.current = null; + }, [documentSettings.language, setDocumentLanguage]); const [parsedOverlayEnabled, setParsedOverlayEnabled] = useState(false); const [isAudioCombining] = useState(false); const audiobookAdapter = useMemo(() => createPdfAudiobookSourceAdapter({ diff --git a/src/app/api/audiobook/chapter/route.ts b/src/app/api/audiobook/chapter/route.ts index 70e78f7..31f40e6 100644 --- a/src/app/api/audiobook/chapter/route.ts +++ b/src/app/api/audiobook/chapter/route.ts @@ -432,7 +432,8 @@ export async function POST(request: NextRequest) { normalizedExistingSettings.nativeSpeed !== normalizedIncomingSettings.nativeSpeed || normalizedExistingSettings.postSpeed !== normalizedIncomingSettings.postSpeed || normalizedExistingSettings.format !== normalizedIncomingSettings.format || - (normalizedExistingSettings.ttsInstructions || '') !== (normalizedIncomingSettings.ttsInstructions || ''); + (normalizedExistingSettings.ttsInstructions || '') !== (normalizedIncomingSettings.ttsInstructions || '') || + (normalizedExistingSettings.language || '') !== (normalizedIncomingSettings.language || ''); if (mismatch) { return NextResponse.json({ error: 'Audiobook settings mismatch', settings: normalizedExistingSettings }, { status: 409 }); } diff --git a/src/app/api/tts/segments/ensure/route.ts b/src/app/api/tts/segments/ensure/route.ts index fdd6b2d..e892784 100644 --- a/src/app/api/tts/segments/ensure/route.ts +++ b/src/app/api/tts/segments/ensure/route.ts @@ -33,6 +33,7 @@ import { getUpstreamRetryAfterSeconds, getUpstreamStatus } from '@/lib/server/tt import { userWhisperAlignJob } from '@/lib/server/jobs/user-whisper-align-job'; import { getResolvedRuntimeConfig } from '@/lib/server/runtime-config'; import { resolveTtsModelForProvider } from '@/lib/shared/tts-provider-policy'; +import { normalizeLanguageTag } from '@/lib/shared/language'; import { resolveSegmentAudioUrls } from '@/lib/server/tts/segment-audio-urls'; import { createRequestLogger, errorToLog } from '@/lib/server/logger'; import { errorResponse } from '@/lib/server/errors/next-response'; @@ -62,6 +63,8 @@ function parseSettings(value: unknown): TTSSegmentSettings | null { if (typeof rec.voice !== 'string') return null; if (!Number.isFinite(Number(rec.nativeSpeed))) return null; if (rec.ttsInstructions !== undefined && typeof rec.ttsInstructions !== 'string') return null; + if (rec.language !== undefined && typeof rec.language !== 'string') return null; + if (typeof rec.language === 'string' && rec.language.length > 64) return null; return { providerRef: rec.providerRef, @@ -70,6 +73,7 @@ function parseSettings(value: unknown): TTSSegmentSettings | null { voice: rec.voice, nativeSpeed: Number(rec.nativeSpeed), ...(typeof rec.ttsInstructions === 'string' ? { ttsInstructions: rec.ttsInstructions } : {}), + ...(typeof rec.language === 'string' ? { language: normalizeLanguageTag(rec.language) } : {}), }; } @@ -406,6 +410,7 @@ export async function POST(request: NextRequest) { alignment = await userWhisperAlignJob({ audioObjectKey: existing.audioKey, text: segment.text, + lang: effectiveSettings.language, sentenceIndex: segment.original.segmentIndex, }); stageTimings.selfHealAlignMs = Date.now() - alignStartedAt; @@ -674,6 +679,7 @@ export async function POST(request: NextRequest) { alignment = await userWhisperAlignJob({ audioObjectKey: audioKey, text: segment.text, + lang: effectiveSettings.language, sentenceIndex: segment.original.segmentIndex, }); stageTimings.whisperAlignMs = Date.now() - alignStartedAt; diff --git a/src/app/api/tts/segments/manifest/route.ts b/src/app/api/tts/segments/manifest/route.ts index c78e602..686555c 100644 --- a/src/app/api/tts/segments/manifest/route.ts +++ b/src/app/api/tts/segments/manifest/route.ts @@ -71,9 +71,10 @@ function parseSettingsValue(value: unknown): TTSSegmentSettings | null { const nativeSpeed = Number.isFinite(Number(speedSource)) ? Number(speedSource) : 1; const instructionsSource = rec.ttsInstructions ?? rec.instructions; const ttsInstructions = typeof instructionsSource === 'string' ? instructionsSource : ''; + const language = typeof rec.language === 'string' ? rec.language : 'en'; if (!providerRef || !providerType || !ttsModel || !voice) return null; - return { providerRef, providerType, ttsModel, voice, nativeSpeed, ttsInstructions }; + return { providerRef, providerType, ttsModel, voice, nativeSpeed, ttsInstructions, language }; } function locatorFromProjection(row: ManifestGroupRow): TTSSegmentLocator | null { diff --git a/src/components/AudiobookExportModal.tsx b/src/components/AudiobookExportModal.tsx index 0bd98b5..9595c46 100644 --- a/src/components/AudiobookExportModal.tsx +++ b/src/components/AudiobookExportModal.tsx @@ -12,6 +12,7 @@ import { useTTS } from '@/contexts/TTSContext'; import { VoicesControlBase } from '@/components/player/VoicesControlBase'; import { ReaderSidebarShell } from '@/components/reader/ReaderSidebarShell'; import { resolveTtsProviderModelPolicy } from '@/lib/shared/tts-provider-policy'; +import { resolveTtsLanguage } from '@/lib/shared/language'; import type { TTSAudiobookChapter, TTSAudiobookFormat } from '@/types/tts'; import { Button, Card, IconButton, MenuActionItem, MenuItemsSurface, RangeInput, SharedListboxButton, SharedListboxOption, SharedListboxOptions } from '@/components/ui'; import { @@ -50,7 +51,7 @@ export function AudiobookExportModal({ onRegenerateChapter }: AudiobookExportModalProps) { const { isLoading, isDBReady, providerRef, providerType, ttsModel, ttsInstructions, voice: configVoice, voiceSpeed, audioPlayerSpeed } = useConfig(); - const { availableVoices } = useTTS(); + const { availableVoices, documentLanguage } = useTTS(); const { progress, setProgress, estimatedTimeRemaining } = useTimeEstimation(); const [isGenerating, setIsGenerating] = useState(false); const [chapters, setChapters] = useState([]); @@ -125,8 +126,12 @@ export function AudiobookExportModal({ postSpeed, format, ttsInstructions: providerModelPolicy.supportsInstructions ? ttsInstructions : undefined, + language: resolveTtsLanguage({ + configuredLanguage: documentLanguage, + voice: nextVoice, + }), }; - }, [savedSettings, audiobookVoice, configVoice, availableVoices, providerRef, providerType, ttsModel, ttsInstructions, effectiveNativeSpeed, postSpeed, format, providerModelPolicy.supportsInstructions]); + }, [savedSettings, audiobookVoice, configVoice, availableVoices, providerRef, providerType, ttsModel, ttsInstructions, effectiveNativeSpeed, postSpeed, format, providerModelPolicy.supportsInstructions, documentLanguage]); const fetchExistingChapters = useCallback(async (soft: boolean = false) => { if (soft) { diff --git a/src/components/documents/DocumentSettings.tsx b/src/components/documents/DocumentSettings.tsx index 07442c2..0e1b0ca 100644 --- a/src/components/documents/DocumentSettings.tsx +++ b/src/components/documents/DocumentSettings.tsx @@ -50,6 +50,20 @@ const viewTypeTextMapping = [ { id: 'scroll', name: 'Continuous Scroll' }, ]; +const DOCUMENT_LANGUAGE_OPTIONS = [ + { value: 'auto', label: 'Automatic (voice or metadata)' }, + { value: 'en', label: 'English' }, + { value: 'es', label: 'Spanish' }, + { value: 'fr', label: 'French' }, + { value: 'hi', label: 'Hindi' }, + { value: 'it', label: 'Italian' }, + { value: 'ja', label: 'Japanese' }, + { value: 'pt-BR', label: 'Portuguese (Brazil)' }, + { value: 'zh-CN', label: 'Chinese (Simplified)' }, + { value: 'ar', label: 'Arabic' }, + { value: 'th', label: 'Thai' }, +]; + type RangeSettingProps = { label: string; value: number; @@ -92,11 +106,13 @@ function RangeSetting({ ); } -export function DocumentSettings({ isOpen, setIsOpen, epub, html, pdf }: { +export function DocumentSettings({ isOpen, setIsOpen, epub, html, language, onLanguageChange, pdf }: { isOpen: boolean, setIsOpen: (isOpen: boolean) => void, epub?: boolean, html?: boolean, + language?: string, + onLanguageChange?: (language: string) => void, pdf?: { parseStatus: PdfParseStatus | null; parsedOverlayEnabled: boolean; @@ -151,6 +167,30 @@ export function DocumentSettings({ isOpen, setIsOpen, epub, html, pdf }: { panelClassName="w-full sm:w-[30rem]" >
+ {language && onLanguageChange ? ( +
+ +
+ ) : null} {isPdfMode && pdf && (
void; setAudioPlayerSpeedAndRestart: (speed: number) => void; setVoiceAndRestart: (voice: string) => void; + documentLanguage: string; + resolvedLanguage: string; + setDocumentLanguage: (language: string) => void; clearSegmentCaches: () => void; skipToLocation: (location: TTSLocation, shouldPause?: boolean) => void; registerLocationChangeHandler: (handler: ((location: TTSLocation) => void) | null) => void; // EPUB-only: Handles chapter navigation @@ -234,8 +238,9 @@ const normalizeLocationKey = normalizeTtsLocationKey; const normalizeBlockFingerprint = (text: string): string => { const normalized = preprocessSentenceForAudio(text) + .normalize('NFKC') .toLowerCase() - .replace(/[^a-z0-9\s]/g, ' ') + .replace(/[^\p{L}\p{N}\p{M}\s]/gu, ' ') .replace(/\s+/g, ' ') .trim(); return normalized.slice(0, 200); @@ -249,6 +254,7 @@ const buildCacheKey = ( model: string, providerType: string, instructions: string, + language: string, ) => { return [ `provider=${provider || ''}`, @@ -257,6 +263,7 @@ const buildCacheKey = ( `voice=${voice || ''}`, `speed=${Number.isFinite(speed) ? speed : ''}`, `instructions=${instructions || ''}`, + `language=${language || ''}`, `text=${sentence}`, ].join('|'); }; @@ -271,10 +278,11 @@ const buildScopedSegmentCacheKey = ( model: string, providerType: string, instructions: string, + language: string, segmentKey?: string | null, ) => { return [ - buildCacheKey(sentence, voice, speed, provider, model, providerType, instructions), + buildCacheKey(sentence, voice, speed, provider, model, providerType, instructions, language), `segmentKey=${segmentKey || ''}`, `locator=${segmentKey ? '' : buildLocatorRequestKey(locator)}`, `segmentIndex=${segmentKey ? '' : segmentIndex}`, @@ -567,6 +575,11 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement const [voice, setVoice] = useState(configVoice); const [ttsModel, setTTSModel] = useState(configTTSModel); const [ttsInstructions, setTTSInstructions] = useState(configTTSInstructions); + const [documentLanguage, setDocumentLanguage] = useState('auto'); + const resolvedLanguage = useMemo( + () => resolveTtsLanguage({ configuredLanguage: documentLanguage, voice }), + [documentLanguage, voice], + ); const providerModelPolicy = useMemo( () => resolveTtsProviderModelPolicy({ providerRef: configProviderRef, @@ -1231,6 +1244,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement maxBlockLength: ttsSegmentMaxBlockLength, keyPrefix: buildSegmentKeyPrefix(documentId, activeReaderType), enforceSourceBoundaries: activeReaderType === 'pdf' && currentUnits !== null && currentUnits.length > 0, + language: resolvedLanguage, }); const currentSegments = plan.segments.filter((segment) => currentSourceKeySet.has(segment.ownerSourceKey)); const newSentences = currentSegments.map((segment) => segment.text); @@ -1383,6 +1397,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement currDocPage, documentId, ttsSegmentMaxBlockLength, + resolvedLanguage, clearPendingEpubJump, ]); @@ -1496,6 +1511,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement voice, effectiveNativeSpeed, providerModelPolicy.supportsInstructions ? ttsInstructions : '', + resolvedLanguage, ttsSegmentMaxBlockLength, ].join('|'); @@ -1516,6 +1532,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement effectiveNativeSpeed, providerModelPolicy.supportsInstructions, ttsInstructions, + resolvedLanguage, ttsSegmentMaxBlockLength, clearPendingEpubJump, bumpEpubPreloadGeneration, @@ -1589,6 +1606,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement ttsModel, configProviderType, providerModelPolicy.supportsInstructions ? ttsInstructions : '', + resolvedLanguage, segmentKey, ); @@ -1669,6 +1687,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement voice, nativeSpeed: effectiveNativeSpeed, ...(providerModelPolicy.supportsInstructions && ttsInstructions ? { ttsInstructions } : {}), + language: resolvedLanguage, }, segments: persistSegments, }, reqHeaders, controller.signal), @@ -1772,6 +1791,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement effectiveNativeSpeed, ttsModel, ttsInstructions, + resolvedLanguage, openApiKey, openApiBaseUrl, configProviderRef, @@ -2253,6 +2273,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement ttsModel, configProviderType, providerModelPolicy.supportsInstructions ? ttsInstructions : '', + resolvedLanguage, playbackSegment?.key, ); const cachedAlignment = sentenceAlignmentCacheRef.current.get(alignmentKey); @@ -2290,6 +2311,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement configProviderType, providerModelPolicy.supportsInstructions, ttsInstructions, + resolvedLanguage, isEPUB, currDocPage, currDocPageNumber, @@ -2388,6 +2410,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement ttsModel, configProviderType, providerModelPolicy.supportsInstructions ? ttsInstructions : '', + resolvedLanguage, nextSegment?.key, ); if (segmentManifestCacheRef.current.has(cacheKey)) { @@ -2498,6 +2521,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement readerType: 'epub', maxBlockLength: ttsSegmentMaxBlockLength, keyPrefix: buildSegmentKeyPrefix(documentId, 'epub'), + language: resolvedLanguage, }); const uniqueCandidates: Array = []; const seenCandidates = new Set(); @@ -2520,6 +2544,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement ttsModel, configProviderType, providerModelPolicy.supportsInstructions ? ttsInstructions : '', + resolvedLanguage, segment.key, ); if (seenCandidates.has(requestKey)) continue; @@ -2564,6 +2589,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement voice, nativeSpeed: effectiveNativeSpeed, ...(providerModelPolicy.supportsInstructions && ttsInstructions ? { ttsInstructions } : {}), + language: resolvedLanguage, }, segments: persistPayload, }, reqHeaders, controller.signal), @@ -2660,6 +2686,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement ttsModel, configProviderType, providerModelPolicy.supportsInstructions ? ttsInstructions : '', + resolvedLanguage, plannedSegment?.key, ); const requestKey = buildSegmentRequestKey(locator, sentenceIndex, sentence, plannedSegment?.key); @@ -2691,6 +2718,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement ttsModel, configProviderType, providerModelPolicy.supportsInstructions ? ttsInstructions : '', + resolvedLanguage, segment.key, ); const requestKey = buildSegmentRequestKey(segmentLocator, index, sentence, segment.key); @@ -2755,6 +2783,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement voice, nativeSpeed: effectiveNativeSpeed, ...(providerModelPolicy.supportsInstructions && ttsInstructions ? { ttsInstructions } : {}), + language: resolvedLanguage, }, segments: persistPayload, }, reqHeaders, controller.signal), @@ -2841,6 +2870,7 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement openApiBaseUrl, providerModelPolicy.supportsInstructions, ttsInstructions, + resolvedLanguage, onTTSStart, onTTSComplete, processSentence, @@ -3145,6 +3175,9 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement setSpeedAndRestart, setAudioPlayerSpeedAndRestart, setVoiceAndRestart, + documentLanguage, + resolvedLanguage, + setDocumentLanguage, clearSegmentCaches, skipToLocation, registerLocationChangeHandler, @@ -3176,6 +3209,8 @@ export function TTSProvider({ children }: { children: ReactNode }): ReactElement setSpeedAndRestart, setAudioPlayerSpeedAndRestart, setVoiceAndRestart, + documentLanguage, + resolvedLanguage, clearSegmentCaches, skipToLocation, registerLocationChangeHandler, diff --git a/src/hooks/epub/useEPUBHighlighting.ts b/src/hooks/epub/useEPUBHighlighting.ts index 8255074..8c31d34 100644 --- a/src/hooks/epub/useEPUBHighlighting.ts +++ b/src/hooks/epub/useEPUBHighlighting.ts @@ -30,6 +30,7 @@ type UseEpubHighlightingParams = { currentWordHighlightCfiRef: MutableRefObject; renderedTextMapsRef: MutableRefObject; wordHighlightMapCacheRef: MutableRefObject; + language?: string; }; type UseEpubHighlightingResult = { @@ -52,6 +53,7 @@ export function useEPUBHighlighting({ currentWordHighlightCfiRef, renderedTextMapsRef, wordHighlightMapCacheRef, + language, }: UseEpubHighlightingParams): UseEpubHighlightingResult { const clearWordHighlights = useCallback(() => { if (!renditionRef.current) return; @@ -116,9 +118,9 @@ export function useEPUBHighlighting({ const resolved = resolveVisibleSegmentRange(renderedTextMapsRef.current, segment); if (!resolved || segment.startAnchor.sourceKey !== resolved.map.sourceKey) return; - const cacheKey = buildWordHighlightCacheKey(segment, alignment); + const cacheKey = buildWordHighlightCacheKey(segment, alignment, language); if (wordHighlightMapCacheRef.current?.key !== cacheKey) { - const tokens = tokenizeCanonicalSegment(segment); + const tokens = tokenizeCanonicalSegment(segment, language); wordHighlightMapCacheRef.current = { key: cacheKey, tokens, @@ -162,6 +164,7 @@ export function useEPUBHighlighting({ renderedTextMapsRef, renditionRef, wordHighlightMapCacheRef, + language, ]); const setRenderedTextMaps = useCallback((maps: EpubRenderedTextMap[]) => { diff --git a/src/hooks/useDocumentLanguage.ts b/src/hooks/useDocumentLanguage.ts new file mode 100644 index 0000000..f8474ab --- /dev/null +++ b/src/hooks/useDocumentLanguage.ts @@ -0,0 +1,52 @@ +'use client'; + +import { useCallback, useEffect, useState } from 'react'; + +import { getDocumentSettings, putDocumentSettings } from '@/lib/client/api/documents'; +import { mergeDocumentSettings } from '@/lib/shared/document-settings'; +import { DEFAULT_DOCUMENT_SETTINGS, type DocumentSettings } from '@/types/document-settings'; + +export function useDocumentLanguage(documentId: string | undefined): { + language: string; + updateLanguage: (language: string) => Promise; +} { + const [settings, setSettings] = useState(DEFAULT_DOCUMENT_SETTINGS); + + useEffect(() => { + setSettings(DEFAULT_DOCUMENT_SETTINGS); + if (!documentId) return; + + const controller = new AbortController(); + void getDocumentSettings(documentId, { signal: controller.signal }) + .then((response) => { + setSettings(mergeDocumentSettings(DEFAULT_DOCUMENT_SETTINGS, response.settings)); + }) + .catch((error) => { + if (error instanceof DOMException && error.name === 'AbortError') return; + console.warn('Failed to load document language, using automatic detection:', error); + }); + + return () => controller.abort(); + }, [documentId]); + + const updateLanguage = useCallback(async (language: string): Promise => { + if (!documentId) return; + const next = mergeDocumentSettings(DEFAULT_DOCUMENT_SETTINGS, { + ...settings, + schemaVersion: 1, + language, + }); + setSettings(next); + try { + const response = await putDocumentSettings(documentId, next); + setSettings(mergeDocumentSettings(DEFAULT_DOCUMENT_SETTINGS, response.settings)); + } catch (error) { + console.warn('Failed to persist document language:', error); + } + }, [documentId, settings]); + + return { + language: settings.language ?? 'auto', + updateLanguage, + }; +} diff --git a/src/lib/client/audiobooks/pipeline.ts b/src/lib/client/audiobooks/pipeline.ts index e9e8675..d2e7c4e 100644 --- a/src/lib/client/audiobooks/pipeline.ts +++ b/src/lib/client/audiobooks/pipeline.ts @@ -12,6 +12,7 @@ import type { TTSAudiobookChapter, TTSAudiobookFormat, } from '@/types/tts'; +import { normalizeTextForTts } from '@/lib/shared/nlp'; export interface PreparedAudiobookChapter { index: number; @@ -109,7 +110,10 @@ export async function runAudiobookGeneration({ maxDelay: 300, }, }: RunAudiobookGenerationOptions): Promise { - const chapters = await adapter.prepareChapters(); + const chapters = (await adapter.prepareChapters()).map((chapter) => ({ + ...chapter, + text: normalizeTextForTts(chapter.text, { language: settings?.language }), + })); const totalLength = chapters.reduce((sum, chapter) => sum + chapter.text.trim().length, 0); if (totalLength === 0) { throw new Error(adapter.noContentMessage); @@ -228,7 +232,7 @@ export async function regenerateAudiobookChapter({ }, }: RegenerateAudiobookChapterOptions): Promise { const chapter = await adapter.prepareChapter(chapterIndex); - const trimmedText = chapter.text.trim(); + const trimmedText = normalizeTextForTts(chapter.text, { language: settings?.language }).trim(); if (!trimmedText) { throw new Error(adapter.noContentMessage); } diff --git a/src/lib/client/epub/epub-word-highlight.ts b/src/lib/client/epub/epub-word-highlight.ts index a249b7e..5e90cda 100644 --- a/src/lib/client/epub/epub-word-highlight.ts +++ b/src/lib/client/epub/epub-word-highlight.ts @@ -1,5 +1,6 @@ import type { CanonicalTtsSegment } from '@/lib/shared/tts-segment-plan'; import type { TTSSentenceAlignment } from '@/types/tts'; +import { normalizeUnicodeToken, segmentWords } from '@/lib/shared/language'; export type EpubCanonicalWordToken = { norm: string; @@ -8,35 +9,19 @@ export type EpubCanonicalWordToken = { }; export const normalizeWordForHighlight = (text: string): string => - text - .toLowerCase() - .replace(/[^a-z0-9]+/g, ''); + normalizeUnicodeToken(text); -export const tokenizeCanonicalSegment = (segment: CanonicalTtsSegment): EpubCanonicalWordToken[] => { - const tokens: EpubCanonicalWordToken[] = []; - const wordRegex = /\S+/g; - let match: RegExpExecArray | null; - - while ((match = wordRegex.exec(segment.text)) !== null) { - const raw = match[0]; - const leading = raw.match(/^[^A-Za-z0-9]*/)?.[0].length ?? 0; - const trailing = raw.match(/[^A-Za-z0-9]*$/)?.[0].length ?? 0; - const start = match.index + leading; - const end = match.index + raw.length - trailing; - if (end <= start) continue; - - const norm = normalizeWordForHighlight(raw.slice(leading, raw.length - trailing)); - if (!norm) continue; - - tokens.push({ - norm, - sourceStart: segment.startAnchor.offset + start, - sourceEnd: segment.startAnchor.offset + end, - }); - } - - return tokens; -}; +export const tokenizeCanonicalSegment = ( + segment: CanonicalTtsSegment, + language?: string, +): EpubCanonicalWordToken[] => + segmentWords(segment.text, language) + .map((token) => ({ + norm: normalizeWordForHighlight(token.text), + sourceStart: segment.startAnchor.offset + token.start, + sourceEnd: segment.startAnchor.offset + token.end, + })) + .filter((token) => Boolean(token.norm)); export const buildMonotonicWordToTokenMap = ( alignmentWords: TTSSentenceAlignment['words'], @@ -105,10 +90,12 @@ export const buildMonotonicWordToTokenMap = ( export const buildWordHighlightCacheKey = ( segment: CanonicalTtsSegment, alignment: TTSSentenceAlignment, + language?: string, ): string => [ segment.key, segment.text.length, + language || '', alignment.words.length, alignment.words.map((word) => normalizeWordForHighlight(word.text)).join('|'), ].join('::'); diff --git a/src/lib/client/html/highlight.ts b/src/lib/client/html/highlight.ts index 104765d..db26cc2 100644 --- a/src/lib/client/html/highlight.ts +++ b/src/lib/client/html/highlight.ts @@ -18,6 +18,7 @@ */ import { CmpStr } from 'cmpstr'; import type { TTSSentenceAlignment } from '@/types/tts'; +import { normalizeUnicodeToken, segmentWords } from '@/lib/shared/language'; export const HTML_SENTENCE_CLASS = 'openreader-html-highlight-sentence'; export const HTML_WORD_CLASS = 'openreader-html-highlight-word'; @@ -53,21 +54,11 @@ interface SentenceState { let sentenceState: SentenceState | null = null; function normalizeWord(word: string): string { - return word - .toLowerCase() - .replace(/[\p{P}\p{S}]+/gu, '') - .trim(); + return normalizeUnicodeToken(word); } -function tokenizePattern(pattern: string): string[] { - const out: string[] = []; - const wordRe = /\S+/g; - let m: RegExpExecArray | null; - while ((m = wordRe.exec(pattern)) !== null) { - const norm = normalizeWord(m[0]); - if (norm) out.push(norm); - } - return out; +function tokenizePattern(pattern: string, language?: string): string[] { + return segmentWords(pattern, language).map((token) => normalizeWord(token.text)).filter(Boolean); } function unwrap(span: HTMLSpanElement): void { @@ -108,7 +99,11 @@ function isHighlightWrapper(node: Node | null): node is HTMLSpanElement { return el.classList.contains(HTML_SENTENCE_CLASS) || el.classList.contains(HTML_WORD_CLASS); } -function collectDomTokens(root: HTMLElement, opts: { skipHighlightWraps: boolean } = { skipHighlightWraps: true }): DomToken[] { +function collectDomTokens( + root: HTMLElement, + language?: string, + opts: { skipHighlightWraps: boolean } = { skipHighlightWraps: true }, +): DomToken[] { const walker = document.createTreeWalker(root, NodeFilter.SHOW_TEXT, { acceptNode(node) { const parent = node.parentElement; @@ -131,15 +126,13 @@ function collectDomTokens(root: HTMLElement, opts: { skipHighlightWraps: boolean while (current) { const textNode = current as Text; const text = textNode.nodeValue || ''; - const wordRe = /\S+/g; - let m: RegExpExecArray | null; - while ((m = wordRe.exec(text)) !== null) { - const norm = normalizeWord(m[0]); + for (const token of segmentWords(text, language)) { + const norm = normalizeWord(token.text); if (!norm) continue; tokens.push({ textNode, - startOffset: m.index, - endOffset: m.index + m[0].length, + startOffset: token.start, + endOffset: token.end, norm, }); } @@ -153,7 +146,7 @@ function collectDomTokens(root: HTMLElement, opts: { skipHighlightWraps: boolean * wrap is applied; lets us index just the words *within* the highlighted * sentence rather than the whole document). */ -function collectTokensInsideWraps(wraps: HTMLSpanElement[]): DomToken[] { +function collectTokensInsideWraps(wraps: HTMLSpanElement[], language?: string): DomToken[] { const tokens: DomToken[] = []; for (const wrap of wraps) { const walker = document.createTreeWalker(wrap, NodeFilter.SHOW_TEXT); @@ -161,15 +154,13 @@ function collectTokensInsideWraps(wraps: HTMLSpanElement[]): DomToken[] { while (current) { const t = current as Text; const text = t.nodeValue || ''; - const wordRe = /\S+/g; - let m: RegExpExecArray | null; - while ((m = wordRe.exec(text)) !== null) { - const norm = normalizeWord(m[0]); + for (const token of segmentWords(text, language)) { + const norm = normalizeWord(token.text); if (!norm) continue; tokens.push({ textNode: t, - startOffset: m.index, - endOffset: m.index + m[0].length, + startOffset: token.start, + endOffset: token.end, norm, }); } @@ -260,14 +251,15 @@ function wrapTokenRange(tokens: DomToken[], start: number, end: number, classNam export function highlightHtmlSentence( container: HTMLElement | null | undefined, sentence: string | null | undefined, + language?: string, ): boolean { clearHtmlSentenceHighlight(); if (!container || !sentence?.trim()) return false; - const patternTokens = tokenizePattern(sentence); + const patternTokens = tokenizePattern(sentence, language); if (!patternTokens.length) return false; - const domTokens = collectDomTokens(container); + const domTokens = collectDomTokens(container, language); if (!domTokens.length) return false; const win = findBestWindow(domTokens, patternTokens); @@ -280,7 +272,7 @@ export function highlightHtmlSentence( // can look up individual word tokens without re-walking the doc. sentenceState = { sentence, - wordTokens: collectTokensInsideWraps(sentenceWraps), + wordTokens: collectTokensInsideWraps(sentenceWraps, language), alignment: null, wordToToken: null, }; diff --git a/src/lib/client/pdf.ts b/src/lib/client/pdf.ts index 9727d4b..32a1a7e 100644 --- a/src/lib/client/pdf.ts +++ b/src/lib/client/pdf.ts @@ -5,6 +5,7 @@ import type { TTSSentenceAlignment } from '@/types/tts'; import type { ParsedPdfDocument, ParsedPdfPage } from '@/types/parsed-pdf'; import { CmpStr } from 'cmpstr'; import type { TTSSegmentLocator } from '@/types/client'; +import { normalizeUnicodeToken, segmentWords } from '@/lib/shared/language'; const cmp = CmpStr.create().setMetric('dice').setFlags('itw'); @@ -181,10 +182,7 @@ function getOrCreateHighlightLayer(span: HTMLElement): { } const normalizeWordForMatch = (text: string): string => - text - .trim() - .replace(/^[^A-Za-z0-9]+|[^A-Za-z0-9]+$/g, '') - .toLowerCase(); + normalizeUnicodeToken(text); // Highlighting functions let highlightPatternSeq = 0; @@ -193,6 +191,7 @@ type HighlightPatternOptions = { parsedDocument?: ParsedPdfDocument | null; locator?: TTSSegmentLocator | null; useBlockGeometryOnly?: boolean; + language?: string; }; function getHighlightLayerForPage(pageElement: HTMLElement): { @@ -451,17 +450,13 @@ export function highlightPattern( const textNode = node as Text; const textContent = textNode.textContent || ''; - const wordRegex = /\S+/g; - let match: RegExpExecArray | null; - - while ((match = wordRegex.exec(textContent)) !== null) { - const word = match[0]; + for (const token of segmentWords(textContent, options?.language)) { tokens.push({ spanIndex, textNode, - text: word, - startOffset: match.index, - endOffset: match.index + word.length, + text: token.text, + startOffset: token.start, + endOffset: token.end, }); } }); diff --git a/src/lib/server/audiobooks/settings.ts b/src/lib/server/audiobooks/settings.ts index cf74673..cdef976 100644 --- a/src/lib/server/audiobooks/settings.ts +++ b/src/lib/server/audiobooks/settings.ts @@ -9,6 +9,7 @@ import type { AudiobookGenerationSettings } from '@/types/client'; import type { TTSAudiobookFormat } from '@/types/tts'; import { resolveEffectiveTtsInstructions } from '@/lib/server/admin/tts-instructions'; import { resolvePreferredSharedProviderSlug } from '@/lib/shared/shared-provider-selection'; +import { normalizeLanguageTag } from '@/lib/shared/language'; function isAudiobookFormat(value: unknown): value is TTSAudiobookFormat { return value === 'mp3' || value === 'm4b'; @@ -68,6 +69,7 @@ export function coerceAudiobookGenerationSettings( postSpeed, format, ...(typeof record.ttsInstructions === 'string' ? { ttsInstructions: record.ttsInstructions } : {}), + ...(typeof record.language === 'string' ? { language: normalizeLanguageTag(record.language) } : {}), }; const migrated = diff --git a/src/lib/server/tts/segments.ts b/src/lib/server/tts/segments.ts index d1bb3c7..e3618ae 100644 --- a/src/lib/server/tts/segments.ts +++ b/src/lib/server/tts/segments.ts @@ -3,6 +3,7 @@ import { mkdtemp, rm, writeFile } from 'fs/promises'; import { join } from 'path'; import { tmpdir } from 'os'; import { preprocessSentenceForAudio } from '@/lib/shared/nlp'; +import { normalizeUnicodeToken, segmentWords } from '@/lib/shared/language'; import { locatorIdentityKey } from '@/lib/shared/tts-locator'; import { ffprobeAudio } from '@/lib/server/audiobooks/chapters'; import type { @@ -32,6 +33,7 @@ function settingsCanonical(settings: TTSSegmentSettings): string { voice: settings.voice, speed: Number.isFinite(Number(settings.nativeSpeed)) ? Number(settings.nativeSpeed) : 1, instructions: settings.ttsInstructions || '', + language: settings.language || 'en', format: 'mp3', }); } @@ -48,6 +50,7 @@ export function buildTtsSegmentSettingsJson(settings: TTSSegmentSettings): TTSSe voice: settings.voice, nativeSpeed: Number.isFinite(Number(settings.nativeSpeed)) ? Number(settings.nativeSpeed) : 1, ttsInstructions: settings.ttsInstructions || '', + language: settings.language || 'en', }; // Postgres jsonb accepts the object directly; SQLite text needs a canonical JSON string. return process.env.POSTGRES_URL ? canonical : settingsCanonical(settings); @@ -242,35 +245,24 @@ export async function probeAudioDurationMsFromBuffer(buffer: Buffer, signal?: Ab } } -function alignWordsToText(sentence: string): Array<{ text: string; charStart: number; charEnd: number }> { - const words = sentence.match(/\S+/g) || []; - const aligned: Array<{ text: string; charStart: number; charEnd: number }> = []; - let cursor = 0; - const lowerSentence = sentence.toLowerCase(); - - for (const token of words) { - const clean = token.trim(); - if (!clean) continue; - const idx = lowerSentence.indexOf(clean.toLowerCase(), cursor); - const start = idx >= 0 ? idx : cursor; - const end = Math.min(sentence.length, start + clean.length); - cursor = Math.max(cursor, end); - aligned.push({ - text: clean, - charStart: start, - charEnd: end, - }); - } - - return aligned; +function alignWordsToText( + sentence: string, + language?: string, +): Array<{ text: string; charStart: number; charEnd: number }> { + return segmentWords(sentence, language).map((token) => ({ + text: token.text, + charStart: token.start, + charEnd: token.end, + })); } export function buildProportionalAlignment(input: { sentence: string; sentenceIndex: number; durationMs: number; + language?: string; }): TTSSentenceAlignment { - const wordsWithOffsets = alignWordsToText(input.sentence); + const wordsWithOffsets = alignWordsToText(input.sentence, input.language); if (wordsWithOffsets.length === 0 || input.durationMs <= 0) { return { sentence: input.sentence, @@ -281,7 +273,7 @@ export function buildProportionalAlignment(input: { const weighted = wordsWithOffsets.map((word) => ({ ...word, - weight: Math.max(1, word.text.replace(/[^a-zA-Z0-9]/g, '').length), + weight: Math.max(1, normalizeUnicodeToken(word.text).length), })); const totalWeight = weighted.reduce((sum, word) => sum + word.weight, 0); diff --git a/src/lib/shared/document-settings.ts b/src/lib/shared/document-settings.ts index 26b45e0..208f76c 100644 --- a/src/lib/shared/document-settings.ts +++ b/src/lib/shared/document-settings.ts @@ -3,6 +3,7 @@ import { type DocumentSettings, } from '@/types/document-settings'; import { PARSED_PDF_BLOCK_KINDS, type ParsedPdfBlockKind } from '@/types/parsed-pdf'; +import { normalizeLanguageTag } from '@/lib/shared/language'; function normalizeSkipKinds(value: unknown): ParsedPdfBlockKind[] { if (!Array.isArray(value)) return [...(DEFAULT_DOCUMENT_SETTINGS.pdf?.skipBlockKinds ?? [])]; @@ -22,6 +23,7 @@ export function mergeDocumentSettings( ): DocumentSettings { const base: DocumentSettings = { schemaVersion: 1, + language: defaults.language || 'auto', pdf: { skipBlockKinds: [...(defaults.pdf?.skipBlockKinds ?? [])], }, @@ -29,12 +31,17 @@ export function mergeDocumentSettings( if (!stored || typeof stored !== 'object') return base; const rec = stored as Record; + const rawLanguage = typeof rec.language === 'string' ? rec.language.trim() : ''; + const language = !rawLanguage || rawLanguage.toLowerCase() === 'auto' + ? 'auto' + : normalizeLanguageTag(rawLanguage, defaults.language || 'en'); const pdf = rec.pdf; - if (!pdf || typeof pdf !== 'object') return base; + if (!pdf || typeof pdf !== 'object') return { ...base, language }; const pdfRec = pdf as Record; return { schemaVersion: 1, + language, pdf: { skipBlockKinds: normalizeSkipKinds(pdfRec.skipBlockKinds), }, diff --git a/src/lib/shared/language.ts b/src/lib/shared/language.ts new file mode 100644 index 0000000..b2e3437 --- /dev/null +++ b/src/lib/shared/language.ts @@ -0,0 +1,119 @@ +export const DEFAULT_TTS_LANGUAGE = 'en'; + +export interface TextToken { + text: string; + start: number; + end: number; +} + +const KOKORO_LANGUAGE_BY_PREFIX: Readonly> = { + af: 'en-US', + am: 'en-US', + bf: 'en-GB', + bm: 'en-GB', + ef: 'es', + em: 'es', + ff: 'fr', + hf: 'hi', + hm: 'hi', + if: 'it', + im: 'it', + jf: 'ja', + jm: 'ja', + pf: 'pt-BR', + pm: 'pt-BR', + zf: 'zh-CN', + zm: 'zh-CN', +}; + +export function normalizeLanguageTag( + language: string | null | undefined, + fallback = DEFAULT_TTS_LANGUAGE, +): string { + const candidate = language?.trim(); + if (!candidate || candidate.toLowerCase() === 'auto') return fallback; + + try { + return Intl.getCanonicalLocales(candidate)[0] ?? fallback; + } catch { + return fallback; + } +} + +export function toBaseLanguageCode(language: string | null | undefined): string { + const normalized = normalizeLanguageTag(language); + try { + return new Intl.Locale(normalized).language; + } catch { + return normalized.split('-')[0]?.toLowerCase() || DEFAULT_TTS_LANGUAGE; + } +} + +export function inferKokoroLanguageFromVoice(voice: string | null | undefined): string | null { + if (!voice?.trim()) return null; + + const languages = new Set( + voice + .split('+') + .map((part) => part.trim().replace(/\([^)]*\)/g, '')) + .map((name) => KOKORO_LANGUAGE_BY_PREFIX[name.slice(0, 2).toLowerCase()]) + .filter((language): language is string => Boolean(language)), + ); + + return languages.size === 1 ? [...languages][0] : null; +} + +export function resolveTtsLanguage(input: { + configuredLanguage?: string | null; + voice?: string | null; +}): string { + const configured = input.configuredLanguage?.trim(); + if (configured && configured.toLowerCase() !== 'auto') { + return normalizeLanguageTag(configured); + } + + return inferKokoroLanguageFromVoice(input.voice) ?? DEFAULT_TTS_LANGUAGE; +} + +export function normalizeUnicodeToken(text: string): string { + return text + .normalize('NFKC') + .toLocaleLowerCase() + .replace(/[^\p{L}\p{N}\p{M}]+/gu, ''); +} + +export function segmentSentences(text: string, language?: string | null): string[] { + const normalizedLanguage = normalizeLanguageTag(language); + try { + return [...new Intl.Segmenter(normalizedLanguage, { granularity: 'sentence' }).segment(text)] + .map(({ segment }) => segment.trim()) + .filter(Boolean); + } catch { + return text.split(/(?<=[.!?。！？؟।])\s*/u).map((segment) => segment.trim()).filter(Boolean); + } +} + +export function segmentWords(text: string, language?: string | null): TextToken[] { + const normalizedLanguage = normalizeLanguageTag(language); + try { + return [...new Intl.Segmenter(normalizedLanguage, { granularity: 'word' }).segment(text)] + .filter((segment) => segment.isWordLike) + .map((segment) => ({ + text: segment.segment, + start: segment.index, + end: segment.index + segment.segment.length, + })); + } catch { + const tokens: TextToken[] = []; + const wordRegex = /\S+/gu; + let match: RegExpExecArray | null; + while ((match = wordRegex.exec(text)) !== null) { + tokens.push({ + text: match[0], + start: match.index, + end: match.index + match[0].length, + }); + } + return tokens; + } +} diff --git a/src/lib/shared/nlp.ts b/src/lib/shared/nlp.ts index b254102..6712116 100644 --- a/src/lib/shared/nlp.ts +++ b/src/lib/shared/nlp.ts @@ -5,13 +5,14 @@ * It handles text preprocessing, sentence splitting, and block creation for optimal TTS processing. */ -import nlp from 'compromise'; +import { normalizeLanguageTag, segmentSentences, toBaseLanguageCode } from '@/lib/shared/language'; export const MAX_BLOCK_LENGTH = 450; const MIN_BLOCK_LENGTH = 50; export interface TtsSplitOptions { maxBlockLength?: number; + language?: string; } function resolveMaxBlockLength(options?: TtsSplitOptions): number { @@ -27,9 +28,9 @@ const splitOversizedText = (text: string, maxLen: number): string[] => { const parts: string[] = []; const MAX_OVERFLOW = maxLen; // allow finishing the sentence up to +maxLen chars - const CLOSERS = new Set(['"', "'", '”', '’', ')', ']', '}']); - const BREAK_CHARS = new Set(['.', '!', '?']); - const SOFT_BREAK_CHARS = new Set([';', ':']); + const CLOSERS = new Set(['"', "'", '”', '’', ')', ']', '}', '」', '』', '】']); + const BREAK_CHARS = new Set(['.', '!', '?', '。', '！', '？', '؟', '।']); + const SOFT_BREAK_CHARS = new Set([';', ':', '；', '：']); const findPunctuationCut = (s: string, limit: number): number | null => { for (let i = limit; i >= 0; i--) { @@ -48,7 +49,7 @@ const splitOversizedText = (text: string, maxLen: number): string[] => { // Allow a boundary at end/whitespace, or common PDF artifact where // the next sentence starts immediately with an uppercase letter. - if (!after || /\s/.test(after) || /[A-Z]/.test(after)) return end; + if (!after || /\s/u.test(after) || /\p{Lu}/u.test(after)) return end; } return null; }; @@ -74,7 +75,7 @@ const splitOversizedText = (text: string, maxLen: number): string[] => { while (cut < s.length && CLOSERS.has(s[cut])) cut++; const after = cut < s.length ? s[cut] : ''; - if (!after || /\s/.test(after) || /[A-Z]/.test(after)) return cut; + if (!after || /\s/u.test(after) || /\p{Lu}/u.test(after)) return cut; } return null; }; @@ -87,7 +88,7 @@ const splitOversizedText = (text: string, maxLen: number): string[] => { let end = i + 1; while (end < s.length && CLOSERS.has(s[end])) end++; const after = end < s.length ? s[end] : ''; - if (!after || /\s/.test(after) || /[A-Z]/.test(after)) return end; + if (!after || /\s/u.test(after) || /\p{Lu}/u.test(after)) return end; } return null; }; @@ -123,8 +124,8 @@ const normalizeSentenceBoundariesForNlp = (text: string): string => { // Insert a space only when it looks like a sentence boundary (lower/digit before, // uppercase after) to avoid breaking abbreviations like "U.S.A". return text - .replace(/([a-z0-9])([.!?])(?=[A-Z])/g, '$1$2 ') - .replace(/([a-z0-9][.!?][\"”’)\]])(?=[A-Z])/g, '$1 '); + .replace(/([\p{Ll}\p{N}])([.!?])(?=\p{Lu})/gu, '$1$2 ') + .replace(/([\p{Ll}\p{N}][.!?]["”’)\]])(?=\p{Lu})/gu, '$1 '); }; /** @@ -136,7 +137,7 @@ const normalizeSentenceBoundariesForNlp = (text: string): string => { export const preprocessSentenceForAudio = (text: string): string => { return text .replace(/\S*(?:https?:\/\/|www\.)([^\/\s]+)(?:\/\S*)?/gi, '- (link to $1) -') - .replace(/(\w+)-\s+(\w+)/g, '$1$2') // Remove hyphenation + .replace(/([\p{L}\p{N}\p{M}]+)-\s+([\p{L}\p{N}\p{M}]+)/gu, '$1$2') // Remove hyphenation // Remove special character * .replace(/\*/g, '') .replace(/\s+/g, ' ') @@ -151,6 +152,8 @@ export const preprocessSentenceForAudio = (text: string): string => { */ export const splitTextToTtsBlocks = (text: string, options?: TtsSplitOptions): string[] => { const maxBlockLength = resolveMaxBlockLength(options); + const language = normalizeLanguageTag(options?.language); + const blockSeparator = ['ja', 'zh'].includes(toBaseLanguageCode(language)) ? '' : ' '; // Treat double-newlines as paragraph boundaries; single newlines are usually // just PDF line wrapping and should not force sentence/block boundaries. const paragraphs = text.split(/\n{2,}/); @@ -162,8 +165,7 @@ export const splitTextToTtsBlocks = (text: string, options?: TtsSplitOptions): s const cleanedText = normalizeSentenceBoundariesForNlp( preprocessSentenceForAudio(paragraph) ); - const doc = nlp(cleanedText); - const rawSentences = doc.sentences().out('array') as string[]; + const rawSentences = segmentSentences(cleanedText, language); // Merge multi-sentence dialogue enclosed in quotes into single items const mergedSentences = mergeQuotedDialogue(rawSentences); @@ -179,8 +181,8 @@ export const splitTextToTtsBlocks = (text: string, options?: TtsSplitOptions): s blocks.push(currentBlock.trim()); currentBlock = sentencePart; } else { - currentBlock = currentBlock - ? `${currentBlock} ${sentencePart}` + currentBlock = currentBlock + ? `${currentBlock}${blockSeparator}${sentencePart}` : sentencePart; } } @@ -200,6 +202,8 @@ export const splitTextToTtsBlocks = (text: string, options?: TtsSplitOptions): s */ export const splitTextToTtsBlocksEPUB = (text: string, options?: TtsSplitOptions): string[] => { const maxBlockLength = resolveMaxBlockLength(options); + const language = normalizeLanguageTag(options?.language); + const blockSeparator = ['ja', 'zh'].includes(toBaseLanguageCode(language)) ? '' : ' '; const paragraphs = text.split(/\n+/); const blocks: string[] = []; @@ -207,8 +211,7 @@ export const splitTextToTtsBlocksEPUB = (text: string, options?: TtsSplitOptions if (!paragraph.trim()) continue; const cleanedText = preprocessSentenceForAudio(paragraph); - const doc = nlp(cleanedText); - const rawSentences = doc.sentences().out('array') as string[]; + const rawSentences = segmentSentences(cleanedText, language); const mergedSentences = mergeQuotedDialogue(rawSentences); @@ -227,7 +230,7 @@ export const splitTextToTtsBlocksEPUB = (text: string, options?: TtsSplitOptions currentBlock = sentencePart; } else { currentBlock = currentBlock - ? `${currentBlock} ${sentencePart}` + ? `${currentBlock}${blockSeparator}${sentencePart}` : sentencePart; } } @@ -248,8 +251,11 @@ export const splitTextToTtsBlocksEPUB = (text: string, options?: TtsSplitOptions * @param {string} text - The text to process * @returns {string} Normalized text */ -export const normalizeTextForTts = (text: string, options?: TtsSplitOptions): string => - splitTextToTtsBlocks(text, options).join(' '); +export const normalizeTextForTts = (text: string, options?: TtsSplitOptions): string => { + const language = normalizeLanguageTag(options?.language); + const separator = ['ja', 'zh'].includes(toBaseLanguageCode(language)) ? '' : ' '; + return splitTextToTtsBlocks(text, options).join(separator); +}; // Helper functions to merge quoted dialogue across sentences const countDoubleQuotes = (s: string): number => { @@ -265,8 +271,8 @@ const countNonApostropheSingleQuotes = (s: string): number => { if (ch === "'" || ch === '‘' || ch === '’') { const prev = i > 0 ? s[i - 1] : ''; const next = i + 1 < s.length ? s[i + 1] : ''; - const isPrevAlphaNum = /[A-Za-z0-9]/.test(prev); - const isNextAlphaNum = /[A-Za-z0-9]/.test(next); + const isPrevAlphaNum = /[\p{L}\p{N}\p{M}]/u.test(prev); + const isNextAlphaNum = /[\p{L}\p{N}\p{M}]/u.test(next); // Treat as a real quote mark only when it's not clearly an apostrophe // between two alphanumeric characters (e.g., don't, WizardLM’s). if (!(isPrevAlphaNum && isNextAlphaNum)) { diff --git a/src/lib/shared/tts-segment-plan.ts b/src/lib/shared/tts-segment-plan.ts index b33ca15..6f76e31 100644 --- a/src/lib/shared/tts-segment-plan.ts +++ b/src/lib/shared/tts-segment-plan.ts @@ -2,7 +2,7 @@ import { preprocessSentenceForAudio, splitTextToTtsBlocks, splitTextToTtsBlocksE import type { TTSSegmentLocator } from '@/types/client'; import type { ReaderType } from '@/types/user-state'; -export const TTS_SEGMENT_PLAN_VERSION = 'tts-segment-plan-v1'; +export const TTS_SEGMENT_PLAN_VERSION = 'tts-segment-plan-v2'; export interface CanonicalTtsSourceUnit { sourceKey: string; @@ -38,6 +38,7 @@ export interface CanonicalTtsSegmentPlanOptions { maxBlockLength?: number; keyPrefix?: string; enforceSourceBoundaries?: boolean; + language?: string; } interface PreparedSourceUnit { @@ -215,7 +216,10 @@ export function planCanonicalTtsSegments( } const canonicalText = textParts.join(''); - const splitOptions = { maxBlockLength: options.maxBlockLength }; + const splitOptions = { + maxBlockLength: options.maxBlockLength, + language: options.language, + }; const splitIntoBlocks = (text: string): string[] => readerType === 'epub' ? splitTextToTtsBlocksEPUB(text, splitOptions) diff --git a/src/types/client.ts b/src/types/client.ts index 0b94363..f9c7440 100644 --- a/src/types/client.ts +++ b/src/types/client.ts @@ -45,6 +45,7 @@ export interface AudiobookGenerationSettings { postSpeed: number; format: TTSAudiobookFormat; ttsInstructions?: string; + language?: string; } export interface CreateChapterPayload { @@ -70,6 +71,7 @@ export interface TTSSegmentSettings { voice: string; nativeSpeed: number; ttsInstructions?: string; + language?: string; } type TTSReaderType = 'pdf' | 'epub' | 'html'; diff --git a/src/types/document-settings.ts b/src/types/document-settings.ts index 6908f33..1d6b358 100644 --- a/src/types/document-settings.ts +++ b/src/types/document-settings.ts @@ -2,6 +2,7 @@ import type { ParsedPdfBlockKind } from '@/types/parsed-pdf'; export interface DocumentSettings { schemaVersion: 1; + language?: string; pdf?: { skipBlockKinds: ParsedPdfBlockKind[]; }; @@ -9,6 +10,7 @@ export interface DocumentSettings { export const DEFAULT_DOCUMENT_SETTINGS: DocumentSettings = { schemaVersion: 1, + language: 'auto', pdf: { skipBlockKinds: ['header', 'footer', 'footnote', 'vision_footnote'], }, diff --git a/tests/files/multilingual-sample.txt b/tests/files/multilingual-sample.txt new file mode 100644 index 0000000..fb517aa --- /dev/null +++ b/tests/files/multilingual-sample.txt @@ -0,0 +1,23 @@ +English +OpenReader should split this sentence correctly. This is the second sentence. + +French +OpenReader doit correctement segmenter cette phrase. Voici la deuxieme phrase. + +Spanish +OpenReader debe segmentar correctamente esta frase. Esta es la segunda frase. + +Arabic +يجب أن يقسم أوبن ريدر هذه الجملة بشكل صحيح. هذه هي الجملة الثانية. + +Hindi +ओपनरीडर को इस वाक्य को सही ढंग से विभाजित करना चाहिए। यह दूसरा वाक्य है। + +Japanese +OpenReaderはこの文を正しく分割する必要があります。これは二番目の文です。 + +Chinese +OpenReader应该正确地分割这个句子。这是第二句话。 + +Thai +OpenReader ควรแบ่งประโยคนี้อย่างถูกต้อง นี่คือประโยคที่สอง diff --git a/tests/unit/audiobook-settings.vitest.spec.ts b/tests/unit/audiobook-settings.vitest.spec.ts index 1cc371c..9cd68fe 100644 --- a/tests/unit/audiobook-settings.vitest.spec.ts +++ b/tests/unit/audiobook-settings.vitest.spec.ts @@ -15,6 +15,7 @@ describe('coerceAudiobookGenerationSettings', () => { postSpeed: 1, format: 'mp3', ttsInstructions: 'keep calm', + language: 'ja-jp', }); expect(result.migrated).toBe(false); @@ -27,6 +28,7 @@ describe('coerceAudiobookGenerationSettings', () => { postSpeed: 1, format: 'mp3', ttsInstructions: 'keep calm', + language: 'ja-JP', }); }); diff --git a/tests/unit/document-settings.vitest.spec.ts b/tests/unit/document-settings.vitest.spec.ts new file mode 100644 index 0000000..9a2256c --- /dev/null +++ b/tests/unit/document-settings.vitest.spec.ts @@ -0,0 +1,22 @@ +import { describe, expect, test } from 'vitest'; + +import { mergeDocumentSettings } from '../../src/lib/shared/document-settings'; +import { DEFAULT_DOCUMENT_SETTINGS } from '../../src/types/document-settings'; + +describe('document settings language', () => { + test('defaults to automatic language resolution', () => { + expect(mergeDocumentSettings(DEFAULT_DOCUMENT_SETTINGS, null).language).toBe('auto'); + }); + + test('normalizes explicit BCP 47 language tags', () => { + expect(mergeDocumentSettings(DEFAULT_DOCUMENT_SETTINGS, { language: 'zh-cn' }).language).toBe('zh-CN'); + expect(mergeDocumentSettings(DEFAULT_DOCUMENT_SETTINGS, { language: 'JA' }).language).toBe('ja'); + }); + + test('preserves language when PDF settings are absent', () => { + expect(mergeDocumentSettings(DEFAULT_DOCUMENT_SETTINGS, { language: 'fr' })).toMatchObject({ + language: 'fr', + pdf: DEFAULT_DOCUMENT_SETTINGS.pdf, + }); + }); +}); diff --git a/tests/unit/epub-word-highlight.vitest.spec.ts b/tests/unit/epub-word-highlight.vitest.spec.ts index 8899dc8..ad1f53c 100644 --- a/tests/unit/epub-word-highlight.vitest.spec.ts +++ b/tests/unit/epub-word-highlight.vitest.spec.ts @@ -28,6 +28,16 @@ const alignmentWords = (words: string[]): TTSSentenceAlignment['words'] => })); describe('EPUB word highlight mapping', () => { + test('tokenizes Japanese and Chinese using locale-aware word boundaries', () => { + const japanese = tokenizeCanonicalSegment(segment('これは日本語です。', 5), 'ja'); + expect(japanese.length).toBeGreaterThan(1); + expect(japanese.every((token) => token.norm.length > 0)).toBe(true); + + const chinese = tokenizeCanonicalSegment(segment('这是中文。', 10), 'zh'); + expect(chinese.length).toBeGreaterThan(1); + expect(chinese.map((token) => token.norm).join('')).toBe('这是中文'); + }); + test('tokenizes canonical segment words with source offsets', () => { const tokens = tokenizeCanonicalSegment(segment('"Hello," she said.', 12)); diff --git a/tests/unit/language.vitest.spec.ts b/tests/unit/language.vitest.spec.ts new file mode 100644 index 0000000..b224a2f --- /dev/null +++ b/tests/unit/language.vitest.spec.ts @@ -0,0 +1,60 @@ +import { describe, expect, test } from 'vitest'; +import { readFileSync } from 'fs'; +import { resolve } from 'path'; + +import { + inferKokoroLanguageFromVoice, + normalizeUnicodeToken, + resolveTtsLanguage, + segmentSentences, + segmentWords, + toBaseLanguageCode, +} from '../../src/lib/shared/language'; + +describe('multilingual language utilities', () => { + test('keeps the multilingual document fixture readable as UTF-8', () => { + const fixture = readFileSync(resolve(process.cwd(), 'tests/files/multilingual-sample.txt'), 'utf8'); + expect(fixture).toContain('これは二番目の文です。'); + expect(fixture).toContain('这是第二句话。'); + expect(fixture).toContain('هذه هي الجملة الثانية.'); + }); + + test('infers Kokoro language from a single-language voice selection', () => { + expect(inferKokoroLanguageFromVoice('jf_alpha')).toBe('ja'); + expect(inferKokoroLanguageFromVoice('zf_xiaobei(0.5)+zm_yunxi(0.5)')).toBe('zh-CN'); + expect(inferKokoroLanguageFromVoice('ff_siwis+jf_alpha')).toBeNull(); + }); + + test('prefers an explicit language and resolves regional tags for Whisper', () => { + expect(resolveTtsLanguage({ configuredLanguage: 'pt-BR', voice: 'jf_alpha' })).toBe('pt-BR'); + expect(resolveTtsLanguage({ configuredLanguage: 'auto', voice: 'jf_alpha' })).toBe('ja'); + expect(toBaseLanguageCode('zh-CN')).toBe('zh'); + }); + + test('normalizes Unicode tokens without dropping non-Latin scripts', () => { + expect(normalizeUnicodeToken('École!')).toBe('école'); + expect(normalizeUnicodeToken('日本語。')).toBe('日本語'); + expect(normalizeUnicodeToken('العربية؟')).toBe('العربية'); + expect(normalizeUnicodeToken('हिन्दी।')).toBe('हिन्दी'); + }); + + test('segments CJK sentences and no-space language words', () => { + expect(segmentSentences('これは最初の文です。これは二番目の文です。', 'ja')).toEqual([ + 'これは最初の文です。', + 'これは二番目の文です。', + ]); + + const chineseWords = segmentWords('这是第一句话。', 'zh').map((token) => token.text); + expect(chineseWords.length).toBeGreaterThan(1); + expect(chineseWords.join('')).toBe('这是第一句话'); + }); + + test('preserves source offsets for Thai word segmentation', () => { + const text = 'นี่คือประโยคแรก'; + const words = segmentWords(text, 'th'); + expect(words.length).toBeGreaterThan(1); + for (const word of words) { + expect(text.slice(word.start, word.end)).toBe(word.text); + } + }); +}); diff --git a/tests/unit/nlp.vitest.spec.ts b/tests/unit/nlp.vitest.spec.ts index 4753b20..b259690 100644 --- a/tests/unit/nlp.vitest.spec.ts +++ b/tests/unit/nlp.vitest.spec.ts @@ -46,6 +46,20 @@ describe('preprocessSentenceForAudio', () => { }); describe('splitTextToTtsBlocks (PDF-oriented)', () => { + test('uses locale-aware sentence boundaries for Japanese and Chinese', () => { + expect(splitTextToTtsBlocks( + 'これは最初の文です。これは二番目の文です。', + { language: 'ja', maxBlockLength: 50 }, + )).toEqual(['これは最初の文です。これは二番目の文です。']); + + const chinese = splitTextToTtsBlocks( + Array(12).fill('这是一个用于测试分句的中文句子。').join(''), + { language: 'zh', maxBlockLength: 50 }, + ); + expect(chinese.length).toBeGreaterThan(1); + expect(chinese.join('')).toContain('用于测试分句'); + }); + test('returns [] for empty input', () => { expect(splitTextToTtsBlocks('')).toEqual([]); expect(splitTextToTtsBlocks(' ')).toEqual([]); @@ -201,4 +215,11 @@ describe('normalizeTextForTts', () => { expect(normalized).not.toMatch(/\s{2,}/); expect(normalized.length).toBeGreaterThan(0); }); + + test('does not insert spaces between normalized Japanese blocks', () => { + expect(normalizeTextForTts('最初の文です。次の文です。', { + language: 'ja', + maxBlockLength: 7, + })).toBe('最初の文です。次の文です。'); + }); }); diff --git a/tests/unit/tts-segments.vitest.spec.ts b/tests/unit/tts-segments.vitest.spec.ts index def8e9f..a448afe 100644 --- a/tests/unit/tts-segments.vitest.spec.ts +++ b/tests/unit/tts-segments.vitest.spec.ts @@ -195,4 +195,19 @@ describe('tts segment helpers', () => { expect(alignment.words[2].endSec).toBeGreaterThan(1.4); expect(alignment.words[1].charStart).toBeGreaterThan(alignment.words[0].charStart); }); + + test('builds proportional alignment for no-space languages', () => { + const sentence = 'これは日本語です'; + const alignment = buildProportionalAlignment({ + sentence, + sentenceIndex: 2, + durationMs: 1200, + language: 'ja', + }); + + expect(alignment.words.length).toBeGreaterThan(1); + for (const word of alignment.words) { + expect(sentence.slice(word.charStart, word.charEnd)).toBe(word.text); + } + }); });