innolux/api/openai_scripts_chinese/audio_processing.py

from openai import OpenAI
from api.openai_scripts_chinese.config import SYSTEM_PROMPT, OPENAI_API_KEY, SUPABASE_KEY, SUPABASE_URL
from supabase import create_client, Client
from api.openai_scripts_chinese.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"請注意以下詞彙：{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"請校對並修正以下轉錄文本，但不要改變其原意或回答問題：\n\n{corrected_transcript}"}
    ]

    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