Update config/settings.py

config/settings.py

@@ -3,8 +3,6 @@ from dotenv import load_dotenv
 
 load_dotenv()
 
-
-
 class Settings:
     GROQ_API_KEY = os.getenv("GROQ_API_KEY")
     
@@ -12,16 +10,17 @@ class Settings:
     VIETNAMESE_EMBEDDING_MODEL = 'keepitreal/vietnamese-sbert'
     VIETNAMESE_LLM_MODEL = "llama-3.1-8b-instant"
     
-    MULTILINGUAL_EMBEDDING_MODEL = 'Qwen/Qwen3-Embedding-0.6B'
+    MULTILINGUAL_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'  # Sử dụng model nhẹ hơn
     MULTILINGUAL_LLM_MODEL = "llama-3.1-8b-instant"
     
     # Fallback models in case primary models fail
     FALLBACK_MULTILINGUAL_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'
     
     # Default models (fallback)
-    DEFAULT_EMBEDDING_MODEL = 'dangvantuan/vietnamese-embedding'
-    DEFAULT_LLM_MODEL = "Vietnamese_LLaMA2_13B_8K_SFT_General_Domain_Knowledge"
+    DEFAULT_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'
+    DEFAULT_LLM_MODEL = "llama-3.1-8b-instant"
     
+    # Whisper Settings - CẬP NHẬT
     WHISPER_MODEL = "whisper-large-v3"
     
     # TTS Settings
@@ -31,17 +30,21 @@ class Settings:
         'de': 'de', 'ja': 'ja', 'ko': 'ko', 'zh': 'zh'
     }
     
-    # RAG Settings
-    EMBEDDING_DIMENSION = 768  # For Vietnamese model
-    MULTILINGUAL_EMBEDDING_DIMENSION = 4096  # For Nemotron model
-    
+    # RAG Settings - CẬP NHẬT
+    EMBEDDING_DIMENSION = 384  # For all-MiniLM-L6-v2
     TOP_K_RESULTS = 3
     
-    # SpeechBrain VAD Settings
-    VAD_MODEL = "speechbrain/vad-crdnn-libriparty"
-    VAD_THRESHOLD = 0.5
-    VAD_MIN_SILENCE_DURATION = 0.8
-    VAD_SPEECH_PAD_DURATION = 0.2
+    # Audio Processing Settings - CẬP NHẬT QUAN TRỌNG
     SAMPLE_RATE = 16000
+    AUDIO_CHUNK_SIZE = 1024
+    AUDIO_SILENCE_THRESHOLD = 0.005  # Ngưỡng âm lượng thấp hơn
+    MIN_AUDIO_DURATION = 0.5  # Giây - âm thanh tối thiểu
+    MAX_AUDIO_DURATION = 10.0  # Giây - âm thanh tối đa
+    
+    # SpeechBrain VAD Settings - CẬP NHẬT
+    VAD_MODEL = "speechbrain/vad-crdnn-libriparty"
+    VAD_THRESHOLD = 0.3  # Giảm threshold để nhạy hơn
+    VAD_MIN_SILENCE_DURATION = 1.0  # Tăng thời gian silence
+    VAD_SPEECH_PAD_DURATION = 0.3   # Tăng padding
 
 settings = Settings()