Better nlp + refactors

Author: richardr1126

src/components/EPUBViewer.tsx

@@ -27,9 +27,10 @@ export function EPUBViewer({ className = '' }: EPUBViewerProps) {
   const rendition = useRef<Rendition | undefined>(undefined);
   const toc = useRef<NavItem[]>([]);
   const locationRef = useRef<string | number>(currDocPage);
+  
 
   const handleLocationChanged = useCallback((location: string | number, initial = false) => {
-    // Handle special 'next' and 'prev' cases
+    // Handle special 'next' and 'prev' cases, which 
     if (location === 'next' && rendition.current) {
       rendition.current.next();
       return;

src/contexts/EPUBContext.tsx

@@ -84,9 +84,7 @@ export function EPUBProvider({ children }: { children: ReactNode }) {
    * Extracts text content from the current EPUB page/location
    */
   const extractPageText = useCallback(async (book: Book, rendition: Rendition): Promise<string> => {
-    try {
-      console.log('Extracting EPUB text from current location');
-      
+    try {      
       const { start, end } = rendition.location;
       if (!start?.cfi || !end?.cfi) return '';
 

src/contexts/TTSContext.tsx

@@ -102,7 +102,7 @@ export function TTSProvider({ children }: { children: React.ReactNode }) {
 
   // Use custom hooks
   const audioContext = useAudioContext();
-  const audioCache = useAudioCache(50);
+  const audioCache = useAudioCache(25);
   const { availableVoices, fetchVoices } = useVoiceManagement(openApiKey, openApiBaseUrl);
 
   // Add ref for location change handler
@@ -120,22 +120,21 @@ export function TTSProvider({ children }: { children: React.ReactNode }) {
    * State Management
    */
   const [isPlaying, setIsPlaying] = useState(false);
+  const [isEPUB, setIsEPUB] = useState(false);
+  const [isProcessing, setIsProcessing] = useState(false);
+
+  const [currDocPage, setCurrDocPage] = useState<string | number>(1);
+  const currDocPageNumber = (!isEPUB ? parseInt(currDocPage.toString()) : 1); // PDF uses numbers only
+  const [currDocPages, setCurrDocPages] = useState<number>();
+
   const [sentences, setSentences] = useState<string[]>([]);
   const [currentIndex, setCurrentIndex] = useState(0);
   const [activeHowl, setActiveHowl] = useState<Howl | null>(null);
-  const [isProcessing, setIsProcessing] = useState(false);
   const [speed, setSpeed] = useState(voiceSpeed);
   const [voice, setVoice] = useState(configVoice);
-  const [currDocPage, setCurrDocPage] = useState<string | number>(1);
-  const [currDocPages, setCurrDocPages] = useState<number>();
   const [nextPageLoading, setNextPageLoading] = useState(false);
 
-  // Add this state to track if we're in EPUB mode
-  const [isEPUB, setIsEPUB] = useState(false);
-
-  const currDocPageNumber = (!isEPUB ? parseInt(currDocPage.toString()) : 1);
-
-  console.log('page:', currDocPage, 'pages:', currDocPages);
+  //console.log('page:', currDocPage, 'pages:', currDocPages);
 
   /**
    * Changes the current page by a specified amount

src/utils/nlp.ts

@@ -15,35 +15,42 @@ export const preprocessSentenceForAudio = (text: string): string => {
     .trim();
 };
 
-const MAX_BLOCK_LENGTH = 250; // Maximum characters per block
+const MAX_BLOCK_LENGTH = 300; // Maximum characters per block
 
 export const splitIntoSentences = (text: string): string[] => {
-  // Preprocess the text before splitting into sentences
-  const cleanedText = preprocessSentenceForAudio(text);
-  const doc = nlp(cleanedText);
-  const rawSentences = doc.sentences().out('array') as string[];
-  
+  // Split text into paragraphs first
+  const paragraphs = text.split(/\n+/);
   const blocks: string[] = [];
-  let currentBlock = '';
 
-  for (const sentence of rawSentences) {
-    const trimmedSentence = sentence.trim();
+  for (const paragraph of paragraphs) {
+    if (!paragraph.trim()) continue;
+
+    // Preprocess each paragraph
+    const cleanedText = preprocessSentenceForAudio(paragraph);
+    const doc = nlp(cleanedText);
+    const rawSentences = doc.sentences().out('array') as string[];
 
-    // If adding this sentence would exceed the limit, start a new block
-    if (currentBlock && (currentBlock.length + trimmedSentence.length + 1) > MAX_BLOCK_LENGTH) {
-      blocks.push(currentBlock.trim());
-      currentBlock = trimmedSentence;
-    } else {
-      // Add to current block with a space if not empty
-      currentBlock = currentBlock 
-        ? `${currentBlock} ${trimmedSentence}`
-        : trimmedSentence;
+    let currentBlock = '';
+
+    for (const sentence of rawSentences) {
+      const trimmedSentence = sentence.trim();
+      
+      // If adding this sentence would exceed the limit, start a new block
+      if (currentBlock && (currentBlock.length + trimmedSentence.length + 1) > MAX_BLOCK_LENGTH) {
+        blocks.push(currentBlock.trim());
+        currentBlock = trimmedSentence;
+      } else {
+        // Add to current block with a space if not empty
+        currentBlock = currentBlock 
+          ? `${currentBlock} ${trimmedSentence}`
+          : trimmedSentence;
+      }
     }
-  }
 
-  // Add the last block if not empty
-  if (currentBlock) {
-    blocks.push(currentBlock.trim());
+    // Add the last block if not empty
+    if (currentBlock) {
+      blocks.push(currentBlock.trim());
+    }
   }
 
   return blocks;

src/utils/pdf.ts

@@ -116,10 +116,12 @@ export function findBestTextMatch(
     lengthDiff: Infinity,
   };
 
+  const SPAN_SEARCH_LIMIT = 10;
+
   for (let i = 0; i < elements.length; i++) {
     let combinedText = '';
     const currentElements = [];
-    for (let j = i; j < Math.min(i + 10, elements.length); j++) {
+    for (let j = i; j < Math.min(i + SPAN_SEARCH_LIMIT, elements.length); j++) {
       const node = elements[j];
       const newText = combinedText ? `${combinedText} ${node.text}` : node.text;
       if (newText.length > maxCombinedLength) break;