Update app.py

app.py

@@ -1,10 +1,9 @@
-from fastapi import FastAPI, HTTPException, WebSocket, WebSocketDisconnect
+from fastapi import FastAPI, WebSocket, WebSocketDisconnect
 from fastapi.middleware.cors import CORSMiddleware
 from gradio_client import Client, handle_file
 import tempfile
 import wave
 import os
-import json
 
 app = FastAPI()
 
@@ -15,46 +14,35 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
-# --- 🧠 延遲連線大腦，防止啟動超時 ---
-brain_cache = {"asr": None, "mt": None}
+# 🧠 懶加載大腦：防止啟動超時
+brain = {"asr": None, "mt": None}
 
 def get_brain(key):
-    if brain_cache[key] is None:
+    if brain[key] is None:
         urls = {
             "asr": "https://ai-labs.ilrdf.org.tw/sapolita-kaldi/",
             "mt": "https://ai-labs.ilrdf.org.tw/kari-seejiq-tnpusu-ai-hmjil/"
         }
-        brain_cache[key] = Client(urls[key])
-    return brain_cache[key]
-
-def parse_dialect(dialect_result):
-    if isinstance(dialect_result, dict) and 'value' in dialect_result:
-        return dialect_result['value']
-    elif isinstance(dialect_result, list) and len(dialect_result) > 0:
-        return dialect_result[0]
-    return dialect_result
+        brain[key] = Client(urls[key])
+    return brain[key]
 
 @app.get("/")
 async def root():
-    return {"status": "online", "mode": "太魯閣語全能模式"}
+    return {"status": "online", "mode": "太魯閣語測試模式"}
 
-# ==========================================
-# 🐉 即時字幕 WebSocket (太魯閣語版本)
-# ==========================================
 @app.websocket("/ws/subtitle")
 async def websocket_subtitle(websocket: WebSocket):
     await websocket.accept()
     audio_buffer = bytearray()
-    ETHNICITY = "太魯閣"
-    DIALECT_ID = "formosan_trv" # ✨ ASR 指定太魯閣語代碼
+    print("📢 太魯閣語專線已啟動")
     
     try:
         while True:
             audio_chunk = await websocket.receive_bytes()
             audio_buffer.extend(audio_chunk)
             
-            # 累積約 3 秒音訊 (260,000 bytes) 提高辨識率
-            if len(audio_buffer) >= 260000:
+            # 累積 25 萬 bytes (約 3 秒) 提高太魯閣語辨識率
+            if len(audio_buffer) >= 250000:
                 with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as temp_wav:
                     with wave.open(temp_wav.name, 'wb') as wav_file:
                         wav_file.setnchannels(1)
@@ -63,38 +51,37 @@ async def websocket_subtitle(websocket: WebSocket):
                         wav_file.writeframes(audio_buffer)
                     temp_path = temp_wav.name
 
-                ind_text, zh_text = "", ""
                 try:
-                    # 1. ASR 辨識 (太魯閣語)
+                    # 1. 太魯閣語 ASR
                     asr_res = get_brain("asr").predict(
-                        dialect_id=DIALECT_ID, 
+                        dialect_id="formosan_trv", 
                         audio_data=handle_file(temp_path), 
                         api_name="/automatic_speech_recognition"
                     )
                     ind_text = str(asr_res).strip()
 
-                    # 2. MT 翻譯 (過濾點點雜音)
-                    if ind_text and ind_text not in [".", "...", "。", ""]:
+                    # 2. 太魯閣語翻譯
+                    zh_text = ""
+                    if ind_text and ind_text not in [".", "...", ""]:
                         mt_brain = get_brain("mt")
-                        # 動態獲取太魯閣語代碼
-                        d_code = parse_dialect(mt_brain.predict(ethnicity=ETHNICITY, api_name="/lambda"))
+                        # 抓取太魯閣語代碼
+                        d_code = "trv_Truku" # 太魯閣語固定代碼
                         zh_res = mt_brain.predict(text=ind_text, src_lang=d_code, tgt_lang="zho_Hant", api_name="/translate")
                         zh_text = str(zh_res).strip()
-                        print(f"🚀 [太魯閣語成功] 族語: {ind_text} | 中文: {zh_text}")
+
+                    await websocket.send_json({
+                        "status": "recognizing",
+                        "indigenous": ind_text,
+                        "chinese": zh_text
+                    })
                 except Exception as e:
-                    print(f"⚠️ 辨識失敗: {e}")
+                    print(f"辨識出錯: {e}")
 
-                await websocket.send_json({
-                    "status": "recognizing",
-                    "indigenous": ind_text if ind_text else "",
-                    "chinese": zh_text
-                })
-                
                 audio_buffer = bytearray()
                 if os.path.exists(temp_path): os.remove(temp_path)
             else:
-                if len(audio_buffer) % 65000 == 0:
-                    await websocket.send_json({"status": "buffering", "text": "🟢 神獸聽太魯閣語中..."})
+                if len(audio_buffer) % 60000 == 0:
+                    await websocket.send_json({"status": "buffering", "text": "🟢 神獸正在聽太魯閣語..."})
     except Exception: pass
 
 if __name__ == "__main__":