from openai import OpenAI
from api.openai_scripts_tai_gi.config import SYSTEM_PROMPT, OPENAI_API_KEY, SUPABASE_KEY, SUPABASE_URL, ERROR_CORRECTION
from supabase import create_client, Client
from api.openai_scripts_tai_gi.text_processing import fuzzy_correct_chinese

client = OpenAI(api_key=OPENAI_API_KEY)
supabase: Client = create_client(SUPABASE_URL, SUPABASE_KEY)

def transcribe(audio_file):
    try:
        table_name = "word_database"
        response = supabase.table(table_name).select("term").execute()
        custom_vocab = []
        if response.data:
            for item in response.data:
                custom_vocab.append({item['term']})
        else:
            print(f"No data found or an error occurred: {response.error}")
            print("Using default dictionary as Supabase data couldn't be fetched.")
        transcript = client.audio.transcriptions.create(
            file=audio_file,
            model="whisper-1",
            response_format="text", 
            prompt=f"""你是一位專業的閩南語轉錄為中文校對助理，專門處理有關長照/日照中心的台語對話轉錄. 
            轉錄時對於「早餐、午餐、晚餐、幼兒園、國小、國中、高中、大學」等詞需特別注意。也需注意以下詞彙：{custom_vocab}"""
        )
        return transcript
    except Exception as e:
        print(f"轉錄時發生錯誤：{str(e)}")
        return None

def post_process_transcript(transcript, temperature=0):
    corrected_transcript = fuzzy_correct_chinese(transcript)
    
    messages = [
        {"role": "system", "content": SYSTEM_PROMPT},
        {"role": "user", "content": f"請校對並修正下面引號內的轉錄文本且只需回傳修正後的文本內容：「{corrected_transcript}」，重點是要查看上述文本是否還有台語的部分，若有則需要修正為繁體中文意思，沒有的話只需要順一下句子。文本裡面若出現「{ERROR_CORRECTION}」這個字典中的字，就直接以此字典內的規則修正文本。最後強調，只需要回傳順過之後的文本，不用加其他不相干的字或是說明。"}
    ]

    response = client.chat.completions.create(
        model="gpt-4",
        temperature=temperature,
        messages=messages
    )

    return response.choices[0].message.content

def process_audio(audio_data):
    raw_transcript = transcribe(audio_data)
    print(raw_transcript)
    if raw_transcript is None:
        return None, None
    corrected_transcript = post_process_transcript(raw_transcript)
    return raw_transcript, corrected_transcript