GameSentenceMiner 2.10.17__py3-none-any.whl → 2.11.0__py3-none-any.whl

GameSentenceMiner/ocr/owocr_helper.py

@@ -195,10 +195,10 @@ all_cords = None
 rectangles = None
 last_ocr2_result = []
 
-def do_second_ocr(ocr1_text, time, img, filtering, ignore_furigana_filter=False):
+def do_second_ocr(ocr1_text, time, img, filtering, ignore_furigana_filter=False, ignore_previous_result=False):
     global twopassocr, ocr2, last_ocr2_result
     try:
-        orig_text, text = run.process_and_write_results(img, None, last_ocr2_result, filtering, None,
+        orig_text, text = run.process_and_write_results(img, None, last_ocr2_result if not ignore_previous_result else None, filtering, None,
                                                         engine=ocr2, furigana_filter_sensitivity=furigana_filter_sensitivity if not ignore_furigana_filter else 0)
 
         if compare_ocr_results(last_ocr2_result, orig_text):
@@ -344,7 +344,8 @@ def run_oneocr(ocr_config: OCRConfig, rectangles):
                 gsm_ocr_config=ocr_config,
                 screen_capture_areas=screen_areas,
                 furigana_filter_sensitivity=furigana_filter_sensitivity,
-                screen_capture_combo=manual_ocr_hotkey if manual_ocr_hotkey and manual else None)
+                screen_capture_combo=manual_ocr_hotkey if manual_ocr_hotkey and manual else None,
+                keep_line_breaks=keep_newline)
     except Exception as e:
         logger.exception(f"Error running OneOCR: {e}")
     done = True
@@ -359,14 +360,14 @@ def add_ss_hotkey(ss_hotkey="ctrl+shift+g"):
     def capture():
         print("Taking screenshot...")
         img = cropper.run()
-        do_second_ocr("", datetime.now(), img, filtering, ignore_furigana_filter=True)
+        do_second_ocr("", datetime.now(), img, filtering, ignore_furigana_filter=True, ignore_previous_result=True)
     def capture_main_monitor():
         print("Taking screenshot of main monitor...")
         with mss.mss() as sct:
             main_monitor = sct.monitors[1] if len(sct.monitors) > 1 else sct.monitors[0]
             img = sct.grab(main_monitor)
             img_bytes = mss.tools.to_png(img.rgb, img.size)
-            do_second_ocr("", datetime.now(), img_bytes, filtering, ignore_furigana_filter=True)
+            do_second_ocr("", datetime.now(), img_bytes, filtering, ignore_furigana_filter=True, ignore_previous_result=True)
     hotkey_reg = None
     try:
         hotkey_reg = keyboard.add_hotkey(ss_hotkey, capture)
@@ -404,7 +405,7 @@ def set_force_stable_hotkey():
 
 if __name__ == "__main__":
     try:
-        global ocr1, ocr2, twopassocr, language, ss_clipboard, ss, ocr_config, furigana_filter_sensitivity, area_select_ocr_hotkey, window, optimize_second_scan, use_window_for_config
+        global ocr1, ocr2, twopassocr, language, ss_clipboard, ss, ocr_config, furigana_filter_sensitivity, area_select_ocr_hotkey, window, optimize_second_scan, use_window_for_config, keep_newline
         import sys
 
         import argparse
@@ -428,6 +429,7 @@ if __name__ == "__main__":
                             help="Optimize second scan by cropping based on first scan results")
         parser.add_argument("--use_window_for_config", action="store_true",
                             help="Use the specified window for loading OCR configuration")
+        parser.add_argument("--keep_newline", action="store_true", help="Keep new lines in OCR output")
 
         args = parser.parse_args()
 
@@ -446,6 +448,7 @@ if __name__ == "__main__":
         clipboard_output = args.clipboard_output
         optimize_second_scan = args.optimize_second_scan
         use_window_for_config = args.use_window_for_config
+        keep_newline = args.keep_newline
 
         window = None
         logger.info(f"Received arguments: {vars(args)}")

GameSentenceMiner/owocr/owocr/ocr.py

@@ -91,8 +91,11 @@ def empty_post_process(text):
     return text
 
 
-def post_process(text):
-    text = ' '.join([''.join(i.split()) for i in text.splitlines()])
+def post_process(text, keep_blank_lines=False):
+    if keep_blank_lines:
+        text = '\n'.join([''.join(i.split()) for i in text.splitlines()])
+    else:
+        text = ''.join([''.join(i.split()) for i in text.splitlines()])
     text = text.replace('…', '...')
     text = re.sub('[・.]{2,}', lambda x: (x.end() - x.start()) * '.', text)
     text = jaconv.h2z(text, ascii=True, digit=True)
@@ -304,22 +307,42 @@ class GoogleLens:
         response_proto = LensOverlayServerResponse().FromString(res.content)
         response_dict = response_proto.to_dict(betterproto.Casing.SNAKE)
 
-        # with open(os.path.join(get_temporary_directory(), 'glens_response.json'), 'w', encoding='utf-8') as f:
-        #     json.dump(response_dict, f, indent=4, ensure_ascii=False)
+        with open(os.path.join(r"C:\Users\Beangate\GSM\Electron App\test", 'glens_response.json'), 'w', encoding='utf-8') as f:
+            json.dump(response_dict, f, indent=4, ensure_ascii=False)
         res = ''
         text = response_dict['objects_response']['text']
         skipped = []
-        if furigana_filter_sensitivity > 0:
-            if 'text_layout' in text:
-                for paragraph in text['text_layout']['paragraphs']:
-                    for line in paragraph['lines']:
+        previous_line = None
+        if 'text_layout' in text:
+            for paragraph in text['text_layout']['paragraphs']:
+                if previous_line:
+                    prev_bbox = previous_line['geometry']['bounding_box']
+                    curr_bbox = paragraph['geometry']['bounding_box']
+                    vertical_space = abs(curr_bbox['center_y'] - prev_bbox['center_y']) * img.height
+                    prev_height = prev_bbox['height'] * img.height
+                    current_height = curr_bbox['height'] * img.height
+                    avg_height = (prev_height + current_height) / 2
+                    # If vertical space is close to previous line's height, add a blank line
+                    # logger.info(f"Vertical space: {vertical_space}, Average height: {avg_height}")
+                    # logger.info(avg_height * 2)
+                    if vertical_space > avg_height * 2:
+                        logger.info('Adding blank line')
+                        res += 'BLANK_LINE'
+                for line in paragraph['lines']:
+                    if furigana_filter_sensitivity:
                         if furigana_filter_sensitivity < line['geometry']['bounding_box']['width'] * img.width and furigana_filter_sensitivity < line['geometry']['bounding_box']['height'] * img.height:
                             for word in line['words']:
                                 res += word['plain_text'] + word['text_separator']
                         else:
                             skipped.append(word['plain_text'] for word in line['words'])
                             continue
-                        res += '\n'
+                    else:
+                        for word in line['words']:
+                                res += word['plain_text'] + word['text_separator']
+                        else:
+                            continue
+                previous_line = paragraph
+                res += '\n'
             # logger.info(
             #     f"Skipped {len(skipped)} chars due to furigana filter sensitivity: {furigana_filter_sensitivity}")
             # widths = []
@@ -350,16 +373,16 @@ class GoogleLens:
             #             else:
             #                 continue
             #         res += '\n'
-        else:
-            if 'text_layout' in text:
-                paragraphs = text['text_layout']['paragraphs']
-                for paragraph in paragraphs:
-                    for line in paragraph['lines']:
-                        for word in line['words']:
-                                res += word['plain_text'] + word['text_separator']
-                        else:
-                            continue
-                    res += '\n'
+        # else:
+        #     if 'text_layout' in text:
+        #         paragraphs = text['text_layout']['paragraphs']
+        #         for paragraph in paragraphs:
+        #             for line in paragraph['lines']:
+        #                 for word in line['words']:
+        #                         res += word['plain_text'] + word['text_separator']
+        #                 else:
+        #                     continue
+        #             res += '\n'
 
         x = (True, res)
 

GameSentenceMiner/owocr/owocr/run.py

@@ -353,7 +353,9 @@ class TextFiltering:
 
         orig_text_filtered = []
         for block in orig_text:
-            if lang == "ja":
+            if "BLANK_LINE" in block:
+                block_filtered = ["\n"]
+            elif lang == "ja":
                 block_filtered = self.kana_kanji_regex.findall(block)
             elif lang == "zh":
                 block_filtered = self.chinese_common_regex.findall(block)
@@ -394,7 +396,8 @@ class TextFiltering:
         new_blocks = []
         for idx, block in enumerate(orig_text):
             if orig_text_filtered[idx] and (orig_text_filtered[idx] not in last_text):
-                new_blocks.append(block)
+                new_blocks.append(str(block).strip().replace("BLANK_LINE", "\n"))
+
 
         final_blocks = []
         if self.accurate_filtering:
@@ -407,9 +410,10 @@ class TextFiltering:
         else:
             for block in new_blocks:
                 # This only filters out NON JA/ZH from text when lang is JA/ZH
-                if lang not in ["ja", "zh"] or self.classify(block)[0] in ['ja', 'zh']:
+                if lang not in ["ja", "zh"] or self.classify(block)[0] in ['ja', 'zh'] or block == "\n":
                     final_blocks.append(block)
 
+
         text = '\n'.join(final_blocks)
         return text, orig_text_filtered
 
@@ -937,7 +941,7 @@ def process_and_write_results(img_or_path, write_to=None, last_result=None, filt
         if filtering:
             text, orig_text = filtering(text, last_result)
         if lang == "ja" or lang == "zh":
-            text = post_process(text)
+            text = post_process(text, keep_blank_lines=keep_new_lines)
         logger.opt(ansi=True).info(f'Text recognized in {end_time - start_time:0.03f}s using <{engine_color}>{engine_instance.readable_name}</{engine_color}>: {text}')
         if notify and config.get_general('notifications'):
             notifier.send(title='owocr', message='Text recognized: ' + text)
@@ -999,6 +1003,7 @@ def run(read_from=None,
         ocr2=None,
         gsm_ocr_config=None,
         furigana_filter_sensitivity=None,
+        keep_line_breaks=False,
         ):
     """
     Japanese OCR client
@@ -1075,11 +1080,13 @@ def run(read_from=None,
     global engine_instances
     global engine_keys
     global lang
+    global keep_new_lines
     lang = language
     engine_instances = []
     config_engines = []
     engine_keys = []
     default_engine = ''
+    keep_new_lines = keep_line_breaks
 
     if len(config.get_general('engines')) > 0:
         for config_engine in config.get_general('engines').split(','):

GameSentenceMiner/util/configuration.py

@@ -16,13 +16,13 @@ import toml
 from dataclasses_json import dataclass_json
 
 OFF = 'OFF'
-VOSK = 'VOSK'
+# VOSK = 'VOSK'
 SILERO = 'SILERO'
 WHISPER = 'WHISPER'
-GROQ = 'GROQ'
+# GROQ = 'GROQ'
 
-VOSK_BASE = 'BASE'
-VOSK_SMALL = 'SMALL'
+# VOSK_BASE = 'BASE'
+# VOSK_SMALL = 'SMALL'
 
 WHISPER_TINY = 'tiny'
 WHISPER_BASE = 'base'
@@ -33,6 +33,7 @@ WHISPER_TURBO = 'turbo'
 
 AI_GEMINI = 'Gemini'
 AI_GROQ = 'Groq'
+AI_LOCAL = 'Local'
 
 INFO = 'INFO'
 DEBUG = 'DEBUG'
@@ -219,7 +220,7 @@ class VAD:
     whisper_model: str = WHISPER_BASE
     do_vad_postprocessing: bool = True
     language: str = 'ja'
-    vosk_url: str = VOSK_BASE
+    # vosk_url: str = VOSK_BASE
     selected_vad_model: str = WHISPER
     backup_vad_model: str = SILERO
     trim_beginning: bool = False
@@ -234,11 +235,11 @@ class VAD:
     def is_whisper(self):
         return self.selected_vad_model == WHISPER or self.backup_vad_model == WHISPER
 
-    def is_vosk(self):
-        return self.selected_vad_model == VOSK or self.backup_vad_model == VOSK
+    # def is_vosk(self):
+    #     return self.selected_vad_model == VOSK or self.backup_vad_model == VOSK
 
-    def is_groq(self):
-        return self.selected_vad_model == GROQ or self.backup_vad_model == GROQ
+    # def is_groq(self):
+    #     return self.selected_vad_model == GROQ or self.backup_vad_model == GROQ
 
 
 @dataclass_json
@@ -266,6 +267,7 @@ class Ai:
     anki_field: str = ''
     provider: str = AI_GEMINI
     gemini_model: str = 'gemini-2.5-flash'
+    local_model: str = OFF
     groq_model: str = 'meta-llama/llama-4-scout-17b-16e-instruct'
     api_key: str = '' # Deprecated
     gemini_api_key: str = ''

GameSentenceMiner/util/text_log.py

@@ -20,6 +20,7 @@ class GameLine:
     next: 'GameLine | None'
     index: int = 0
     scene: str = ""
+    TL: str = ""
 
     def get_previous_time(self):
         if self.prev:
@@ -31,6 +32,9 @@ class GameLine:
             return self.next.time
         return 0
 
+    def set_TL(self, tl: str):
+        self.TL = tl
+
     def __str__(self):
         return str({"text": self.text, "time": self.time})