Update app.py

app.py

@@ -40,31 +40,52 @@ def extract_feature(signal: np.ndarray, sr: int) -> np.ndarray:
 
 # --- 4. 三種預測函式 ---
 
-def predict_face(img: np.ndarray):
-    print("predict_face called, img is None?", img is None)
-    # 你的限频和 DeepFace 分析逻辑...
+def predict_face(img):
+    global _last_time, _last_result
+    if img is None:
+        return {}
+    now = time.time()
+    # 限频: 每 0.5 秒最多分析一次
+    if now - _last_time < 0.5 and _last_result:
+        return _last_result
     try:
-        result = DeepFace.analyze(img, actions=["emotion"], detector_backend="opencv")
-        emo = result.get("emotion", {})
-        print("DeepFace result:", emo)
+        res = DeepFace.analyze(img, actions=["emotion"], detector_backend="opencv")
+        # 处理返回类型
+        if isinstance(res, list):
+            first = res[0] if len(res) > 0 else {}
+            emo = first.get("emotion", {}) if isinstance(first, dict) else {}
+        elif isinstance(res, dict):
+            emo = res.get("emotion", {})
+        else:
+            emo = {}
+        _last_result = emo
+        _last_time = now
+        print("predict_face result:", emo)
         return emo
     except Exception as e:
         print("DeepFace.analyze error:", e)
-        return {}
+        # 出错时返回上次有效结果或空
+        return _last_result if _last_result else {}
 
 
-def predict_voice(audio):
-    """
-    語音情緒分析：audio 由 Gradio 傳入，形式為暫存檔路徑字串 (str)。
-    用 librosa.load 讀取，再提取 MFCC 特徵，最後用 SVM 模型 predict_proba。
-    回傳格式為 dict，例如 {"angry":0.1, "happy":0.7, ...}
-    """
-    # audio 參數為 Gradio Audio 組件給的檔案路徑
-    signal, sr = librosa.load(audio, sr=None)
-    feat = extract_feature(signal, sr)
-    probs = svm_model.predict_proba([feat])[0]
-    labels = svm_model.classes_
-    return {labels[i]: float(probs[i]) for i in range(len(labels))}
+
+def predict_voice(audio_path: str):
+    # 如果没有录音文件路径，直接返回空字典或提示
+    if not audio_path:
+        # 可打印日志，帮助调试
+        print("predict_voice: 收到 None 或空 audio_path，跳過分析")
+        return {}
+    try:
+        signal, sr = librosa.load(audio_path, sr=None)
+        # 提取特征
+        feat = extract_feature(signal, sr)  # 你的特征提取函数
+        probs = svm_model.predict_proba([feat])[0]
+        labels = svm_model.classes_
+        return {labels[i]: float(probs[i]) for i in range(len(labels))}
+    except Exception as e:
+        print("predict_voice error:", e)
+        return {}
+