app.py

import os
from fastapi import FastAPI, UploadFile, File, Body
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import JSONResponse
from fastapi.concurrency import run_in_threadpool
from fastapi import HTTPException
from datetime import datetime
from collections import Counter
import pandas as pd
import torch
import uvicorn
import asyncio
import logging
from config import UPLOAD_FOLDER, SUPABASE_URL, SUPABASE_KEY
from models import pipelines, models, model_lock, load_model_bundle
from utils import (
    save_uploaded_file,
    extract_and_normalize_audio,
    diarize_audio,
    split_segments,
    transcribe_segments,
    clean_summary,
    summarize_texts,
    add_llm_spell_corrected_text_column
)
# from supabase import create_client, Client

logger = logging.getLogger(__name__)

app = FastAPI()
app.router.redirect_slashes = False
origins = [o.strip() for o in (os.getenv("CORS_ORIGINS", "").split(",")) if o.strip()] or ["*"]
app.add_middleware(
    CORSMiddleware,
    allow_origins=origins,
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

# ตั้งค่า Supabase URL และ KEY จาก env
# SUPABASE_URL = os.environ.get("SUPABASE_URL")
# SUPABASE_KEY = os.environ.get("SUPABASE_KEY")
# supabase: Client = create_client(SUPABASE_URL, SUPABASE_KEY)

@app.on_event("startup")
async def startup_event():
    async with model_lock:
        if pipelines and models is not None:
            logger.info("ℹ️ Models already loaded. Skipping startup load.")
            return

        logger.info("🔁 Loading models at startup...")
        try:
            pipeline, model = await load_model_bundle()
        except Exception as e:
            logger.exception(f"❌ Model loading failed: {e}")
            import sys; sys.exit(1)

        # idempotent set (load_model_bundle ก็ทำแล้ว แต่นี่กันไว้เผื่อเรียกจากที่อื่น)
        if not pipelines:
            pipelines.append(pipeline)
        else:
            pipelines[0] = pipeline

        if not models:
            models.append(model)
        else:
            models[0] = model

        logger.info("✅ Models loaded successfully at startup")

@app.get("/")
async def check_api():
    pipelines_loaded = len(pipelines)
    models_loaded = len(models)

    return {
        "status": "running",
        "models_loaded": {
            "pipelines": pipelines_loaded,
            "whisper_models": models_loaded,
            "pipeline_ready": pipelines_loaded > 0 and all(pipelines),
            "model_ready": models_loaded > 0 and all(models),
        },
        "cuda": {
            "available": torch.cuda.is_available(),
            "device_count": torch.cuda.device_count() if torch.cuda.is_available() else 0,
            "device_names": [
                torch.cuda.get_device_name(i) for i in range(torch.cuda.device_count())
            ] if torch.cuda.is_available() else []
        },
        "timestamp": datetime.now().isoformat(),
        "debug": {
        "pipelines": [str(p) for p in pipelines],
        "models": [str(m) for m in models],
},
    }

@app.get("/key")
async def check_env():
    import os
    return {
        "env": os.environ.get("ENV", "dev"),
        "openai_key_exists": bool(os.environ.get("OPENAI_API_KEY")),
    }

# @app.post('/video')
# async def upload_video(file: UploadFile = File(...)):
#     video_path = save_uploaded_file(file)
#     return video_path
@app.api_route("/upload_video", methods=["POST"])
@app.api_route("/upload_video/", methods=["POST"])
async def process_video(file: UploadFile = File(...)):
    try:
        payload = await run_in_threadpool(upload_video, file)
        return JSONResponse(content=payload)
    except Exception as e:
        import traceback, sys
        traceback.print_exc(file=sys.stderr)
        raise HTTPException(status_code=500, detail=f"processing failed: {e}")

def upload_video(file: UploadFile = File(...)):
    import os
    from config import together_api_key
    video_path = save_uploaded_file(file)
    audio_path = extract_and_normalize_audio(video_path)
    df_diarization = diarize_audio(audio_path)
    segment_folder = split_segments(audio_path, df_diarization)
    df_transcriptions = transcribe_segments(segment_folder)
    min_len = min(len(df_diarization), len(df_transcriptions))
    df_merged = pd.concat([
        df_diarization.iloc[:min_len].reset_index(drop=True),
        df_transcriptions.iloc[:min_len].reset_index(drop=True)
    ], axis=1)
    # df_merged = add_corrected_text_column(df_merged)
    df_merged = add_llm_spell_corrected_text_column(df_merged)
    # summaries = summarize_texts(df_merged["text"].tolist(), together_api_key, delay=0)
    result = df_merged.to_dict(orient="records")
    speaker_array = df_diarization["speaker"].unique().tolist()
    counter = Counter(df_diarization["speaker"])
    result_array = [{"speaker": spk, "count": cnt} for spk, cnt in counter.most_common()]
    from pydub import AudioSegment
    duration_minutes = len(AudioSegment.from_wav(audio_path)) / 1000 / 60
    # save result to supabase
    # supabase.table("summaries").insert(result).execute()
    return {
        "video_path": video_path,
        "audio_path": audio_path,
        "audio_length": duration_minutes,
        "data": result,
        "speaker_array": speaker_array,
        "count_speaker": result_array,
        "num_speakers": len(speaker_array),
        "total_sentence": len(df_merged['text']),
        "summaries": 'This feature not available',
    }


@app.post("/sumarize/")
async def sumarize(array: list = Body(...)):
    from config import together_api_key
    import pandas as pd
    df = pd.DataFrame(array)
    summaries = summarize_texts(df["text"].tolist(), together_api_key, delay=2)
    # save to supabase
    data_to_save = [{"text": row["text"], "summary": summary} for row, summary in zip(array, summaries)]
    # supabase.table("summaries").insert(data_to_save).execute()
    return JSONResponse(content={
        "summaries": summaries,
    })

if __name__ == "__main__":
    uvicorn.run(
        app,
        host="0.0.0.0",
        port=7860,
        reload=False
    )