server.py

import os
import time
import shutil
import subprocess
import difflib  # 文字比對用
from fastapi import FastAPI, UploadFile, File, Form
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import JSONResponse
from fastapi.staticfiles import StaticFiles
from gradio_client import Client, handle_file  # ✨ 記得匯入 handle_file

app = FastAPI()

# 允許跨域請求
app.add_middleware(
    CORSMiddleware, allow_origins=["*"], allow_credentials=True, allow_methods=["*"], allow_headers=["*"],
)

UPLOAD_DIR = "temp_uploads"
if not os.path.exists(UPLOAD_DIR): os.makedirs(UPLOAD_DIR)

# 格式為: "帳號名稱/專案名稱"
ASR_API_URL = "ithuan/sapolita-kaldi"

# --- 核心 1: 轉檔 (WebM -> WAV) ---
def convert_webm_to_wav(webm_path):
    wav_path = webm_path.replace(".webm", ".wav")
    
    # 判斷 ffmpeg 指令
    if os.path.exists("ffmpeg.exe"): ffmpeg_cmd = "ffmpeg.exe"
    else: ffmpeg_cmd = "ffmpeg"
    
    # ASR 需要 16kHz 單聲道
    cmd = f'{ffmpeg_cmd} -y -i "{webm_path}" -ar 16000 -ac 1 "{wav_path}" -loglevel quiet'
    
    try:
        subprocess.run(cmd, shell=True, check=True)
        return wav_path
    except:
        print("❌ FFmpeg 轉檔失敗，請確認 ffmpeg.exe 是否存在")
        return None

# --- 核心 2: 呼叫外部 ASR API (依照您提供的新邏輯) ---
def speech_to_text(audio_path):
    print(f"📡 正在傳送音檔至 ASR: {ASR_API_URL} ...")
    
    try:
        client = Client(ASR_API_URL)
        
        # ✨ 設定太魯閣語代碼
        lang_code = "formosan_trv"
        
        # ✨ 呼叫 API: 參數順序是 (語言代碼, 音檔)
        # 並且使用 handle_file 包裝音檔
        result = client.predict(
            lang_code,                  # 參數 1: 語言代碼
            handle_file(audio_path),    # 參數 2: 音檔
            api_name="/automatic_speech_recognition"
        )
        
        # result 通常回傳純文字，或是包含文字的 Tuple
        # 這裡做個轉型確保安全
        text_result = str(result).strip()
        print(f"👂 ASR 聽到: {text_result}")
        return text_result
        
    except Exception as e:
        print(f"❌ ASR API 呼叫失敗: {e}")
        return None

# --- 核心 3: 文字相似度評分 ---
def calculate_text_score(user_text, target_text):
    if not user_text or user_text == "None": return 0
    
    # 1. 統一轉小寫並去除標點 (太魯閣語常有 ? 或 !)
    u = user_text.lower().replace("?", "").replace("!", "").replace(".", "").strip()
    t = target_text.lower().replace("?", "").replace("!", "").replace(".", "").strip()
    
    print(f"🔍 比對文字:\n  - 學生說: {u}\n  - 正確答案: {t}")

    # 2. 完全命中
    if u == t: return 100
    
    # 3. 模糊比對
    similarity = difflib.SequenceMatcher(None, u, t).ratio()
    score = int(similarity * 100)
    
    # 鼓勵機制：只要有開口，且辨識出東西，就加一點分
    if score > 0: score = min(100, score + 10)
    
    return score

# --- API 接口 ---
@app.post("/api/score")
async def score_audio(file: UploadFile = File(...), target_word: str = Form(...)):
    user_path = None
    wav_path = None
    try:
        # 1. 存下使用者錄音
        filename = f"user_{int(time.time())}.webm"
        user_path = os.path.join(UPLOAD_DIR, filename)
        with open(user_path, "wb") as f:
            f.write(await file.read())
        
        # 2. 轉檔
        wav_path = convert_webm_to_wav(user_path)
        if not wav_path:
            return JSONResponse(content={"status": "error", "message": "轉檔失敗"})

        # 3. 呼叫 ASR
        asr_result = speech_to_text(wav_path)
        
        # 4. 評分
        if asr_result:
            real_score = calculate_text_score(asr_result, target_word)
            
            # 依分數給評語
            if real_score == 100: 
                comment = f"太強了！完全正確！(辨識: {asr_result})"
            elif real_score >= 80: 
                comment = f"發音很標準喔！(辨識: {asr_result})"
            elif real_score >= 60: 
                comment = f"很接近了，加油！(辨識: {asr_result})"
            else: 
                comment = f"AI 聽成：{asr_result}，請再清楚一點！"
        else:
            real_score = 0
            comment = "AI 聽不太清楚，請再試一次！"

        return JSONResponse(content={"status": "success", "score": real_score, "comment": comment})

    except Exception as e:
        print(f"Server Error: {str(e)}")
        return JSONResponse(content={"status": "error", "message": str(e)})
    finally:
        # 清理暫存檔
        if user_path and os.path.exists(user_path): 
            try: os.remove(user_path)
            except: pass
        if wav_path and os.path.exists(wav_path): 
            try: os.remove(wav_path)
            except: pass

# --- 啟動網頁伺服器 ---
app.mount("/", StaticFiles(directory=".", html=True), name="static")

if __name__ == "__main__":
    import uvicorn
    print("🚀 ASR (Sapolita-Kaldi) 伺服器啟動中... (http://localhost:8000)")
    uvicorn.run("server:app", host="0.0.0.0", port=8000, reload=True)