[fix]: update.

config.py

@@ -73,3 +73,4 @@ LLM_SYS_PROMPT_EN = """
 """
 
 hotwords_file = MODEL_DIR / 'hotwords.txt'
+hotwords_json = MODEL_DIR / 'hotwords.json'

moyoyo_asr_models/hotwords.json

@@ -1,3 +1,7 @@
 {
-    "勾色母": "GOSIM"
+    "高斯姆": "GOSIM",
+    "GO SIM": "GOSIM",
+    "go sim": "GOSIM",
+    "GO SAME": "GOSIM",
+    "go same": "GOSIM"
 }

moyoyo_asr_models/hotwords.txt

@@ -10,7 +10,6 @@ lib
 HUAWEI
 Future为
 Cloud
-NVIDIA
 OpenAI
 PYTHON
 八币

transcribe/helpers/funasr.py

@@ -21,7 +21,7 @@ class FunASR:
     def warmup(self, warmup_steps=1):
         warmup_soundfile = f"{config.ASSERT_DIR}/jfk.flac"
         for _ in range(warmup_steps):
-            self.model.generate(input=warmup_soundfile, disable_pbar=True, hotword=config.hotwords_file.as_posix())
+            self.model.generate(input=warmup_soundfile, disable_pbar=True)
 
     def transcribe(self, audio_buffer: bytes, language):
         audio_frames = np.frombuffer(audio_buffer, dtype=np.float32)

transcribe/utils.py

@@ -8,6 +8,7 @@ import config
 import csv
 import av
 import re
+import json
 
 # Compile regex patterns once outside the loop for better performance
 p_pattern = re.compile(r"(\s*\[.*?\])")
@@ -18,43 +19,79 @@ p_end_pattern = re.compile(r"(\s*.*\])")
 def filter_words(res_word):
     """
     Filter words according to specific bracket patterns.
-    
+
     Args:
         res_word: Iterable of word objects with a 'text' attribute
-    
+
     Returns:
         List of filtered word objects
     """
     asr_results = []
     skip_word = False
-    
+
     for word in res_word:
         # Skip words that completely match the pattern
         if p_pattern.match(word.text):
             continue
-        
+
         # Mark the start of a section to skip
         if p_start_pattern.match(word.text):
             skip_word = True
             continue
-        
+
         # Mark the end of a section to skip
         if p_end_pattern.match(word.text) and skip_word:
             skip_word = False
             continue
-        
+
         # Skip words if we're in a skip section
         if skip_word:
             continue
-        
+
+        word.text = replace_hotwords(word.text)
+
         # Add the word to results if it passed all filters
         asr_results.append(word)
-    
+
     return asr_results
 
+
+
+def replace_hotwords(text: str) -> str:
+    """
+    Reads hotwords from a JSON file and replaces occurrences in the input text.
+
+    Args:
+        text: The input string to process.
+
+    Returns:
+        The string with hotwords replaced.
+    """
+    # Using the provided absolute path for simplicity in this example
+    hotwords_path = config.hotwords_file.as_posix()
+
+    try:
+        with open(hotwords_path, 'r', encoding='utf-8') as f:
+            hotwords = json.load(f)
+    except FileNotFoundError:
+        print(f"Error: hotwords.json not found at {hotwords_path}")
+        return text
+    except json.JSONDecodeError:
+        print(f"Error: Could not decode JSON from {hotwords_path}")
+        return text
+
+    processed_text = text
+    # Iterate through the hotwords dictionary
+    for key, value in hotwords.items():
+        # Replace all occurrences of the key with the value in the text
+        processed_text = processed_text.replace(key, value)
+
+    return processed_text
+
+
 def log_block(key: str, value, unit=''):
     if config.DEBUG:
-        return 
+        return
     """格式化输出日志内容"""
     key_fmt = f"[  {key.ljust(25)}]"  # 左对齐填充
     val_fmt = f"{value} {unit}".strip()
@@ -171,4 +208,4 @@ class TestDataWriter:
     def write(self, result: 'DebugResult'):
         with open(self.file_path, mode='a', newline='') as file:
             writer = csv.DictWriter(file, fieldnames=self.fieldnames)
-            writer.writerow(result.model_dump(by_alias=True))
+            writer.writerow(result.model_dump(by_alias=True))