Execute Custom JS prior to TTS queue extraction in core.js

[snthsh2]

Describe your suggested feature

Problem Statement

LNReader provides an excellent feature allowing users to inject custom user JavaScript via 'Settings → Reader → Advanced' to alter the layout or clean text. However, when using the Advanced TTS engine, text-scraping configurations inside the background service grab the elements via window.tts.getTextNodes() before the custom user scripts finish executing asynchronously or observing DOM changes.

As a result, aggressive anti-scraper watermarks (such as lookalike text blocks, unicode swaps, and sentence spaces corrupted by strings like u2014 on platforms like Novelight) bypass our custom cleaning scripts and are read out loud by the native Android TTS engine.

Describe the solution you'd like

Instead of hardcoding site-specific cleaning regex into the app repository, I would like to request a structural timing change to how custom user scripts interact with the TTS cycle.

If the WebView core script (core.js) hooks and runs user-defined Custom JS synchronously right at the beginning of the window.tts.getTextNodes execution block, users will have the flexibility to scrub, manipulate, or filter text elements locally before they ever touch the Android TTS background queue.

This keeps the repository lightweight and gives power-users an immediate way to handle complex anti-bot sentence formatting dynamically.

Other details

Implementation Idea / Example

Inside android/app/src/main/assets/js/core.js, we can ensure any user-injected functions or a designated global cleaning hook are triggered right before the text serialization array is built:

window.tts = {
    getTextNodes: function() {
        // 1. Invoke custom reader JS/cleanup configurations first if present
        if (typeof window.executeCustomUserReaderJS === 'function') {
            window.executeCustomUserReaderJS();
        }
        
        // 2. Original array parsing and extraction happens here...
        let paragraphs = [];
        // ...
        return paragraphs;
    }
};

This tiny execution hook gives us a permanent bridge to clean up layout text and keep background audio completely pristine without forcing the core dev team to maintain scraper-fighting scripts!

Acknowledgements

• I have searched the existing issues and this is a new ticket, NOT a duplicate or related to another open or closed issue.
• I have written a short but informative title.
• If this is an issue with a source, I should be opening an issue in the sources repository.
• I have updated the app to version 2.0.2.
• I will fill out all of the requested information in this form.

[snthsh2]

Real-World Example of Use Case Script

Here is the robust, custom optimization script I currently use to clean up text nodes. Because this script relies on heavy typography normalization, it must run synchronously inside window.tts.getTextNodes to prevent the raw watermarks from leaking into the native Android text queue:

// Example of user script intended to run prior to native audio node selection
function cleanNovelightWatermarks() {
  const unicodeMap = {
    '𝐍':'N','𝓝':'N','𝔑':'N','о':'o','𝑜':'o','𝐨':'o',
    'е':'e','𝑒':'e','𝐞':'e','і':'i','ι':'i','𝐢':'i',
    'г':'g','𝐠':'g','һ':'h','𝐡':'h','т':'t','𝐭':'t'
  };

  const connector = /(?:[\W_]|u2014)*/.source;
  const baseNamePattern = `N${connector}o${connector}v${connector}e${connector}[li]${connector}i${connector}g${connector}h${connector}t`;
  const warningPhrases = ['\\(Official version\\)', '\\(Read the full story\\)', '\\(Original source\\)', '\\(Continue reading\\)', 'Do not rehost this novel', `\\(Only on(?:\\s*${baseNamePattern})?\\)`].join('|');

  const watermarkRegex = new RegExp(
    `(?:[\\W_]|u2014)*(?:(?:This(?:[\\W_]|u2014)*translation(?:[\\W_]|u2014)*is(?:[\\W_]|u2014)*the(?:[\\W_]|u2014)*intellectual(?:[\\W_]|u2014)*property(?:[\\W_]|u2014)*of|Do not rehost this novel)\\s*)?` +
    baseNamePattern + `\\s*(?:(?:\\s*(?:${warningPhrases}))|(?:[\\W_]|u2014))*`, 'gi'
  );

  document.querySelectorAll('p, span, a, div').forEach(el => {
    if (el.children.length === 0 && el.textContent) {
      // Step A: Fix string-space corruption ("weeku2014would")
      if (el.textContent.includes('u2014')) {
        el.textContent = el.textContent.replaceAll('u2014', ' ');
      }
      
      // Step B: Sanitize hidden unicode fonts
      let sanitizedText = el.textContent;
      for (let badChar in unicodeMap) {
        sanitizedText = sanitizedText.replaceAll(badChar, unicodeMap[badChar]);
      }
      sanitizedText = sanitizedText.normalize("NFD").replace(/[\u0300-\u036f]/g, "");
      
      // Step C: Slice watermark out with space padding to prevent merged words
      if (watermarkRegex.test(sanitizedText)) {
        if (sanitizedText.replace(watermarkRegex, '').trim().length === 0) {
          el.remove();
        } else {
          watermarkRegex.lastIndex = 0;
          let rawText = el.textContent, offset = 0, match;
          while ((match = watermarkRegex.exec(sanitizedText)) !== null) {
            let start = match.index - offset;
            let end = start + match.length;
            rawText = rawText.substring(0, start) + " " + rawText.substring(end);
            offset += (match.length - 1);
          }
          el.textContent = rawText;
        }
      }
      el.textContent = el.textContent.replace(/ {2,}/g, ' ').trim();
    }
  });
}

Providing an execution window for scripts like this ensures that text processing is completed before the app's native TTS component reads out the raw data.

[snthsh2]

💡 Additional Use Case: Custom Phonetic Pronunciation Maps for TTS

Another major benefit of allowing custom user JavaScript to run synchronously inside window.tts.getTextNodes() is that it enables advanced users to build Phonetic Pronunciation Dictionaries.

Fantasy and translated light novels frequently contain made-up words, unusual localized names, or honorifics that default Android system TTS engines mispronounce terribly. Because the custom hook executes right before the native audio layer pulls the text, users can leverage their scripts to seamlessly swap difficult text strings with their phonetic equivalents for a much more natural-sounding voice narrative:

// Example: Global Phonetic Map for translated Light Novels
const phoneticMap = {
  "Xianxia": "Shee-an-shah",
  "Qing": "Ching",
  "-sama": " sam-ah",
  "-kun": " koon",
  "Ainz": "Ownz"
};

// This quick map can be executed at the end of the text node cleanup loop
let textOutput = el.textContent;
for (let realWord in phoneticMap) {
  textOutput = textOutput.replaceAll(new RegExp(`\\b${realWord}\\b`, 'gi'), phoneticMap[realWord]);
}
el.textContent = textOutput;

This structural hook solves anti-scraper text corruption while simultaneously giving the community a clean way to handle custom pronunciation configurations without bloating the base application settings.

[bizzkoot]

@snthsh2 you can try PR #18 to see if this address the needs