Trixy7/trixy_core/stt/layers/hunspell_layer.py

# -*- coding: utf-8 -*-
"""
Hunspell-Korrekturschicht.

Woerterbuch-basierte Korrektur (~6ms/Wort).
"""

from pathlib import Path

from trixy_core.stt.layers.base import CorrectionLayer
from trixy_core.utils.debug import pdebug, pwarn

try:
    import hunspell
    _HAS_HUNSPELL = True
except ImportError:
    _HAS_HUNSPELL = False


class HunspellLayer(CorrectionLayer):
    """Hunspell woerterbuch-basierte Korrektur."""

    NAME = "hunspell"

    def __init__(self) -> None:
        self._hobj: "hunspell.HunSpell | None" = None
        self._protected_words: set[str] = set()

    def is_available(self) -> bool:
        """Prueft ob pyhunspell installiert ist."""
        return _HAS_HUNSPELL

    def initialize(self, config: dict, language: str, protected_words: list[str]) -> bool:
        """Laedt Hunspell-Woerterbuch."""
        if not _HAS_HUNSPELL:
            return False

        dict_path = config.get("hunspell_dict_path", f"/usr/share/hunspell/{language}_DE")

        # Hunspell braucht .dic und .aff Dateien
        dic_file = Path(f"{dict_path}.dic")
        aff_file = Path(f"{dict_path}.aff")

        if not dic_file.exists() or not aff_file.exists():
            pdebug(f"[Hunspell] Woerterbuch nicht gefunden: {dict_path}")
            return False

        try:
            self._hobj = hunspell.HunSpell(str(aff_file), str(dic_file))
            self._protected_words = {w.lower() for w in protected_words}

            # Geschuetzte Woerter zum Woerterbuch hinzufuegen
            for word in protected_words:
                self._hobj.add(word)

            pdebug(f"[Hunspell] Woerterbuch geladen: {dict_path}")
            return True

        except Exception as e:
            pwarn(f"[Hunspell] Initialisierung fehlgeschlagen: {e}")
            self._hobj = None
            return False

    def correct(self, text: str) -> str:
        """Prueft jedes Wort einzeln und schlaegt Korrekturen vor."""
        if not self._hobj or not text.strip():
            return text

        try:
            words = text.split()
            corrected_words = []
            changed = False

            for word in words:
                # Geschuetzte Woerter ueberspringen
                if word.lower() in self._protected_words:
                    corrected_words.append(word)
                    continue

                # Wort pruefen (Satzzeichen entfernen fuer Pruefung)
                clean_word = word.strip(".,!?;:\"'()")
                if not clean_word:
                    corrected_words.append(word)
                    continue

                if self._hobj.spell(clean_word):
                    # Wort ist korrekt
                    corrected_words.append(word)
                else:
                    # Wort ist falsch — Vorschlaege holen
                    suggestions = self._hobj.suggest(clean_word)
                    if suggestions:
                        # Ersten Vorschlag verwenden (bester Match)
                        suggestion = suggestions[0]
                        # Satzzeichen beibehalten
                        prefix = word[:len(word) - len(word.lstrip(".,!?;:\"'()"))] if word != word.lstrip(".,!?;:\"'()") else ""
                        suffix = word[len(word.rstrip(".,!?;:\"'()")):] if word != word.rstrip(".,!?;:\"'()") else ""
                        corrected_words.append(f"{prefix}{suggestion}{suffix}")
                        changed = True
                    else:
                        corrected_words.append(word)

            if changed:
                result = " ".join(corrected_words)
                pdebug(f"[Hunspell] '{text}' → '{result}'")
                return result

            return text

        except Exception as e:
            pdebug(f"[Hunspell] Fehler bei Korrektur: {e}")
            return text

    def shutdown(self) -> None:
        """Gibt Ressourcen frei."""
        self._hobj = None