ksbyte/normalize/_upstream/normalizer.py

from typing import List, Dict
from .constants import (
    KASHMIRI_CHARACTER_MAPPING,
    KASHMIRI_PUNCTUATIONS, PUNCTUATION_MAP,
    KASHMIRI_ENG_DIGITS_MAP, ENG_KASHMIRI_DIGITS_MAP, 
    KASHMIRI_DIACRITICS, WORD_TO_DIGIT_MAP, ALL_CHARACTERS)
import regex as re

class KashmiriNormalizer:
    def __init__(self):
        """Initialize the normalizer"""
        pass

    def _replace(self, text: str, charMap: Dict[str, List[str]]) -> str:
        """Replaces the letters in the list in map's values with its key"""
        flattenMap: Dict = {}
        for key, value in charMap.items():
            for letter in value:
                flattenMap[letter] = key
                
        if flattenMap:
            # Sort by length (descending) to prevent substring issues
            sorted_bad_chars = sorted(flattenMap.keys(), key=len, reverse=True)
            
            pattern_string = "|".join(map(re.escape, sorted_bad_chars))
            
            regex_pattern = re.compile(pattern_string)
        else:
            regex_pattern = None
            
        if not regex_pattern:
            return text
            
        # The lambda finds the match in the inverted map and returns the correct key
        return regex_pattern.sub(
            lambda match: flattenMap[match.group(0)], 
            text
        )
        
    def _punctuation_spaces(self, text: str) -> str:
        """Removes spaces before punctuations and add spaces after them"""
        
        escaped_puncts = "".join([re.escape(p) for p in KASHMIRI_PUNCTUATIONS])
        
        # Add spaces after punctuations, numbers and some special characters
        _SPACE_AFTER_PUNCTUATIONS_RE = re.compile(
            r"(?<=[" + escaped_puncts + r"])(?=[^" + escaped_puncts + r"0-9 \n])",
            flags=re.U | re.M | re.I)
        
        # Remove whitespaces before punctuations
        _REMOVE_SPACE_BEFORE_PUNCTUATIONS_RE = re.compile(r'\s+([' + escaped_puncts + r'])',
                                                        flags=re.U | re.M | re.I)
        
        text = _SPACE_AFTER_PUNCTUATIONS_RE.sub(' ', text)
        text = _REMOVE_SPACE_BEFORE_PUNCTUATIONS_RE.sub(r'\1', text)
        return text
        
    def _canonicalize(self, text: str) -> str:
        """Canonicalize text using Kashmiri Character maps."""
        
        text = self._replace(text, KASHMIRI_CHARACTER_MAPPING)
        text = self._replace(text, PUNCTUATION_MAP)
        return text

    def _replace_digits(self, text: str, toEnglish: bool = True) -> str:
        """Replaces Kashmiri (Persio-Arabic) Digits with English (Latin) Digits and vice versa"""
        if not toEnglish:
            return self._replace(text, KASHMIRI_ENG_DIGITS_MAP)
        return self._replace(text, ENG_KASHMIRI_DIGITS_MAP)
    # def _handlePlatYe(self, text: str) -> str:
    #     """Replaces ؠ with ۍ when it occurs at the final position of words to align with Kashmiri writing rules"""
        
    #     t = re.escape('ؠ')
    #     pattern = fr"\b{t}|{t}\b"
    #     return re.sub(pattern, "ۍ", text) # Because in Kashmiri writing ؠ changes to ۍ at final position
    
    def _removeDiacritics(self, text: str) -> str:
        """
        Removes all the diacritics from the input text
        NOTE: According to linguists diacritics are important in kashmiri unlike urdu, so don't remove them, this function is just to perform tests and for research purposes
        """
        REP_MAP = {"": list(KASHMIRI_DIACRITICS)}
        return self._replace(text, REP_MAP)
    
    def normalize(self, text: str, removeDiacritics: bool = False) -> str:
        """
        1. Canonicalizes the text
        2. Replaces Kashmiri digits with English 
        3. Handles spaces before and after punctuations
        
        Ideal for Pre-Processing of ML models.
        Args:
            text (str): The input text to normalize.
            removeDiacritics (bool): Do you want to remove Diacritics or not?
            
        Returns:
            str: The normalized text.
        """
        text = self._canonicalize(text)
        text = self._replace_digits(text)
        text = self._punctuation_spaces(text)
        if removeDiacritics:
            text = self._removeDiacritics(text)
        
        return text


class TTSNormalizer(KashmiriNormalizer):
    def __init__(self):
        """Initialize the TTS normalizer."""
        super().__init__()

    def _convert_digits_to_words(self, text: str) -> str:
        """Converts digits to their word forms"""
        return self._replace(text, WORD_TO_DIGIT_MAP)

    def _handlePlatYe(self, text: str) -> str:
        """Replaces ؠ with ۍ when it occurs at the final position of words to align with Kashmiri writing rules"""
        
        t = re.escape('ؠ')
        pattern = fr"\b{t}|{t}\b"
        return re.sub(pattern, "ۍ", text) # Because in Kashmiri writing ؠ changes to ۍ at final position

    def _remove_non_kashmiri_characters(self, text: str) -> str:
        """Removes all the characters that are not in Kashmiri Language"""
        return "".join([char for char in text if char in ALL_CHARACTERS or char == '\n'])

    def normalize(self, text: str) -> str:
        """
        Normalizes text specifically for TTS tasks.
        1. Canonicalizes the text
        2. Handles Plat Ye (ؠ -> ۍ at end of words)
        3. Replaces Kashmiri digits with English digits
        4. Converts English digits to Kashmiri words
        5. Handles spaces before and after punctuations
        6. Preserves diacritics (always)
        
        Args:
            text (str): The input text to normalize.
            
        Returns:
            str: The normalized text.
        """
        text = self._canonicalize(text)
        text = self._handlePlatYe(text)
        text = self._replace_digits(text)
        text = self._convert_digits_to_words(text)
        text = self._punctuation_spaces(text)
        text = self._remove_non_kashmiri_characters(text)
        
        
        return text