config/settings.py

# import os
# from dotenv import load_dotenv

# load_dotenv()

# class Settings:
#     GROQ_API_KEY = os.getenv("GROQ_API_KEY")
    
#     # Multilingual Model Settings
#     VIETNAMESE_EMBEDDING_MODEL = 'keepitreal/vietnamese-sbert'
#     VIETNAMESE_LLM_MODEL = "llama-3.1-8b-instant"
    
#     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 = '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
#     MAX_CHUNK_LENGTH = 200
#     SUPPORTED_LANGUAGES = {
#         'vi': 'vi', 'en': 'en', 'fr': 'fr', 'es': 'es', 
#         'de': 'de', 'ja': 'ja', 'ko': 'ko', 'zh': 'zh'
#     }
    
#     # RAG Settings - CẬP NHẬT
#     EMBEDDING_DIMENSION = 768  # For all-MiniLM-L6-v2
#     TOP_K_RESULTS = 5
    
#     # 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 = "snakers4/silero-vad"
#     VAD_THRESHOLD = 0.2  # Giảm threshold để nhạy hơn
#     VAD_MIN_SILENCE_DURATION = 3  # Tăng thời gian silence
#     VAD_SPEECH_PAD_DURATION = 0.3   # Tăng padding

# settings = Settings()
import os
from dotenv import load_dotenv

load_dotenv()

class Settings:
    GROQ_API_KEY = os.getenv("GROQ_API_KEY")
    
    # Multilingual Model Settings
    VIETNAMESE_EMBEDDING_MODEL = 'keepitreal/vietnamese-sbert'
    VIETNAMESE_LLM_MODEL = "llama-3.1-8b-instant"
    
    MULTILINGUAL_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'
    MULTILINGUAL_LLM_MODEL = "llama-3.1-8b-instant"
    
    # Fallback models
    FALLBACK_MULTILINGUAL_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'
    
    # Default models
    DEFAULT_EMBEDDING_MODEL = 'sentence-transformers/all-MiniLM-L6-v2'
    DEFAULT_LLM_MODEL = "llama-3.1-8b-instant"
    
    # OCR Settings - SỬA LỖI NGÔN NGỮ
    OCR_MODEL = "kha-white/manga-ocr-base"
    EASYOCR_LANGUAGES = ['vi', 'en']  # SỬA 'zh' thành 'ch_sim'
    
    # Whisper Settings
    WHISPER_MODEL = "whisper-large-v3"
    
    # TTS Settings
    MAX_CHUNK_LENGTH = 200
    SUPPORTED_LANGUAGES = {
        'vi': 'vi', 'en': 'en', 'fr': 'fr', 'es': 'es', 
        'de': 'de', 'ja': 'ja', 'ko': 'ko', 'zh': 'zh'
    }
    
    # RAG Settings
    EMBEDDING_DIMENSION = 768
    TOP_K_RESULTS = 5
    
    # Audio Processing Settings
    SAMPLE_RATE = 16000
    AUDIO_CHUNK_SIZE = 1024
    AUDIO_SILENCE_THRESHOLD = 0.003
    MIN_AUDIO_DURATION = 0.8
    MAX_AUDIO_DURATION = 15.0
    
    # VAD Settings
    VAD_MODEL = "snakers4/silero-vad"
    VAD_THRESHOLD = 0.3
    VAD_MIN_SPEECH_DURATION = 1.0
    VAD_MIN_SILENCE_DURATION = 2.0
    VAD_SPEECH_PAD_DURATION = 0.5
    VAD_PRE_SPEECH_BUFFER = 0.3

settings = Settings()