"""
ocr_reader.py
-------------
Wraps EasyOCR so the app can read signs, menus, room numbers, etc.
Loaded lazily/cached because EasyOCR's model init is slow (~2-5s) and
should only happen once per session, not once per frame.
"""

from typing import List, Tuple
import numpy as np
import easyocr


class TextReader:
    def __init__(self, languages: List[str] = None, gpu: bool = False):
        """
        languages: EasyOCR language codes, e.g. ['en'] or ['en', 'fr'].
        gpu: set True if a CUDA GPU is available -- big speedup for OCR.
        """
        self.languages = languages or ["en"]
        self.reader = easyocr.Reader(self.languages, gpu=gpu)

    def read_text(self, frame: np.ndarray, min_confidence: float = 0.4) -> List[Tuple[str, float]]:
        """
        Runs OCR on a single frame.
        Returns a list of (text, confidence) tuples, left-to-right / top-to-bottom
        in the order EasyOCR detects them, filtered by min_confidence.
        """
        raw_results = self.reader.readtext(frame)  # [(bbox, text, confidence), ...]

        filtered = []
        for _, text, confidence in raw_results:
            cleaned = text.strip()
            if cleaned and confidence >= min_confidence:
                filtered.append((cleaned, confidence))

        return filtered

    def read_text_as_sentence(self, frame: np.ndarray, min_confidence: float = 0.4) -> str:
        """Convenience method: joins all detected text fragments into one spoken sentence."""
        results = self.read_text(frame, min_confidence)
        if not results:
            return ""
        return ". ".join(text for text, _ in results)