Update src/streamlit_app.py

src/streamlit_app.py

@@ -2,44 +2,31 @@
 # src/streamlit_app.py
 
 import os
-# ========== 1. 环境准备（修复权限问题） ==========
-# Streamlit 会在 STREAMLIT_HOME 下写文件，强制指向 /app/.streamlit
+# ① 告訴 Streamlit 將設定檔讀自專案的 .streamlit 資料夾
 os.environ["STREAMLIT_HOME"] = os.path.join(os.getcwd(), ".streamlit")
-os.makedirs(os.environ["STREAMLIT_HOME"], exist_ok=True)
 
-# DeepFace 会在 DEEPFACE_HOME 下缓存模型权重，指向 /app/.deepface
-os.environ["DEEPFACE_HOME"] = os.path.join(os.getcwd(), ".deepface")
-os.makedirs(os.environ["DEEPFACE_HOME"], exist_ok=True)
-
-# ========== 2. 引入依赖 ==========
 import streamlit as st
-from streamlit_webrtc import VideoTransformerBase, webrtc_streamer
-import cv2
-import numpy as np
-import librosa
-import joblib
+import cv2, numpy as np, base64, io
+import librosa, joblib
 from deepface import DeepFace
 
-# ========== 3. 模型加载 ==========
-st.set_page_config(page_title="多模態情緒分析", layout="wide")
-
+# —— 1. 預先載入模型 —— 
 @st.cache_resource(show_spinner=False)
-def load_face_warmup():
-    # 预热 DeepFace，避免首次卡顿
+def load_models():
+    # a) 先讓 DeepFace 熱身（不實際偵測人臉）
     DeepFace.analyze(
-        img_path = np.zeros((224,224,3), dtype="uint8"),
-        actions = ["emotion"],
-        enforce_detection = False,
+        img_path = np.zeros((224,224,3), dtype=np.uint8),
+        actions  = ['emotion'],
+        enforce_detection=False
     )
-    return True
+    # b) 載入你本機訓練好的語音模型
+    audio_model = joblib.load("voice_model.joblib")
+    return audio_model
 
-@st.cache_resource(show_spinner="Loading audio model…")
-def load_audio_model():
-    # 必须保证 voice_model.joblib 在 /app 根目录
-    return joblib.load("voice_model.joblib")
+audio_model = load_models()
 
-# 文字情绪简单规则
-def analyze_text_fn(text:str) -> str:
+# —— 2. 文本情緒函式 —— 
+def analyze_text_fn(text):
     if any(w in text for w in ["開心","快樂","愉快","喜悅","歡喜","興奮","歡","高興"]):
         return "happy"
     if any(w in text for w in ["生氣","憤怒","不爽","發火","火大","氣憤"]):
@@ -50,79 +37,43 @@ def analyze_text_fn(text:str) -> str:
         return "surprise"
     if any(w in text for w in ["怕","恐懼","緊張","懼","膽怯","畏"]):
         return "fear"
-    return "neutral"
+    return "neutral" 
 
-# 语音情绪函数
-def analyze_audio_fn(path:str) -> str:
-    y, sr = librosa.load(path, sr=None, duration=3, offset=0.5)
+# —— 3. 語音情緒函式 —— 
+def analyze_audio_fn(wav_bytes):
+    # 將上傳的 bytes 用 librosa 讀入
+    y, sr = librosa.load(io.BytesIO(wav_bytes), sr=None)
+    # 計算 MFCC 並取均值作為特徵
     mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13)
-    feats = np.mean(mfccs.T, axis=0)
-    model = load_audio_model()
-    return model.predict([feats])[0]
-
-# 预加载
-_ = load_face_warmup()
-audio_model = load_audio_model()
+    mf = np.mean(mfccs.T, axis=0)
+    # 回傳模型預測結果
+    return audio_model.predict([mf])[0]
 
-# ========== 4. 前端界面 ==========
-st.title("📱 多模態實時情緒分析（人臉/語音/文字）")
+# —— 4. 網頁介面佈局 —— 
+st.title("📱 多模態即時情緒分析")
 
-tabs = st.tabs(["👤 人臉實時", "🎤 語音上傳", "✍️ 文本分析"])
+tabs = st.tabs([
+    "🔴 臉部（僅限本地）", 
+    "🎤 語音上傳", 
+    "⌨️ 文字輸入"
+])
 
-# ----- Tab1: 人脸实时 -----
 with tabs[0]:
-    st.subheader("實時人臉情緒")
-    class FaceEmotionTransformer(VideoTransformerBase):
-        def __init__(self):
-            self.last_emo = "neutral"
-        def transform(self, frame):
-            img = frame.to_ndarray(format="bgr24")
-            res = DeepFace.analyze(
-                img, actions=["emotion"], enforce_detection=False
-            )
-            # DeepFace 可能返回 list 或 dict
-            if isinstance(res, list):
-                emo = res[0].get("dominant_emotion", "unknown")
-            else:
-                emo = res.get("dominant_emotion", "unknown")
-            self.last_emo = emo
-            # 在图上标文字
-            cv2.putText(img, emo, (10,40),
-                        cv2.FONT_HERSHEY_SIMPLEX, 1.2,
-                        (0,255,0), 2, cv2.LINE_AA)
-            return img
-
-    ctx = webrtc_streamer(
-        key="face-emotion",
-        mode="SENDRECV",
-        media_stream_constraints={"video": True, "audio": False},
-        video_transformer_factory=FaceEmotionTransformer,
-        async_transform=True,
-    )
-
-    if ctx.video_transformer:
-        st.markdown(f"**目前檢測到情緒：** `{ctx.video_transformer.last_emo}`")
-    else:
-        st.write("請點擊下方「Start」按鈕開始攝像頭")
+    st.header("實時臉部（本地瀏覽器測試用）")
+    st.info("⚠️ HF Spaces 無法直接存取攝影機，僅本地測試有效。")
 
-# ----- Tab2: 语音上传 -----
 with tabs[1]:
-    st.subheader("上傳 WAV文件 進行語音情緒分析")
-    audio_file = st.file_uploader("选择 .wav 文件", type=["wav"])
-    if audio_file is not None:
-        tmp_path = "tmp_audio.wav"
-        with open(tmp_path, "wb") as f:
-            f.write(audio_file.getbuffer())
-        emo = analyze_audio_fn(tmp_path)
-        st.success(f"🎤 語音情緒預測：**{emo}**")
-        os.remove(tmp_path)
+    st.header("上傳 WAV 檔案進行分析")
+    wav_file = st.file_uploader("選擇一個 .wav 音訊檔", type="wav")
+    if wav_file:
+        emo = analyze_audio_fn(wav_file.read())
+        st.success(f"🎤 語音檢測到的情緒：**{emo}**")
 
-# ----- Tab3: 文本分析 -----
 with tabs[2]:
-    st.subheader("輸入文字 進行情緒分析")
-    txt = st.text_area("在此輸入中文句子", height=100)
-    if st.button("分析文字情緒"):
+    st.header("輸入文字進行分析")
+    txt = st.text_area("在此輸入或貼上文字")
+    if st.button("開始分析"):
         emo = analyze_text_fn(txt)
-        st.success(f"✍️ 文本情緒預測：**{emo}**")
+        st.success(f"📝 文本檢測到的情緒：**{emo}**")