app.py

# app.py
import os
# ─── 解決 DeepFace 無法寫入預設路徑的問題 ─────────────────────────────────────────
# 將 DeepFace 的快取目錄指向可寫入的 /tmp 之下
os.environ["DEEPFACE_HOME"] = "/tmp/.deepface"

import gradio as gr
import numpy as np
import joblib, io
import librosa
from deepface import DeepFace

# ─── 1. 載入模型 ─────────────────────────────────────────────────────────────
# 我們把模型檔放在跟 app.py 同一層的 voice_model.joblib
MODEL_PATH = os.path.join(os.path.dirname(__file__), "voice_model.joblib")
audio_model = joblib.load(MODEL_PATH)

# ─── 2. 定義各種分析函數 ────────────────────────────────────────────────────
def analyze_face(frame: np.ndarray):
    # DeepFace 回傳 dict，裡面有 'dominant_emotion'
    res = DeepFace.analyze(frame, actions=["emotion"], enforce_detection=False)
    return frame, res["dominant_emotion"]

def analyze_audio(wav_file):
    wav_bytes = wav_file.read()
    # 用 librosa 讀入
    y, sr = librosa.load(io.BytesIO(wav_bytes), sr=None)
    mfccs = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13)
    mf = np.mean(mfccs.T, axis=0)
    return audio_model.predict([mf])[0]

def analyze_text(txt):
    # 簡單關鍵字 mapping
    mapping = {
        "😊 happy":   ["開心","快樂","愉快","喜悅","歡喜","興奮","高興","歡"],
        "😠 angry":   ["生氣","憤怒","不爽","發火","火大","氣憤"],
        "😢 sad":     ["傷心","難過","哭","憂","悲","心酸","哀","痛苦","慘","愁"],
        "😲 surprise":["驚訝","意外","嚇","好奇","驚詫","詫異","訝異"],
        "😨 fear":    ["怕","恐懼","緊張","懼","膽怯","畏"],
    }
    for emo, kws in mapping.items():
        if any(w in txt for w in kws):
            return emo
    return "😐 neutral"

# ─── 3. 建立 Gradio 介面 ────────────────────────────────────────────────────
with gr.Blocks(title="多模態即時情緒分析") as demo:
    gr.Markdown("## 🤖 多模態即時情緒分析")
    with gr.Tabs():
        with gr.TabItem("📷 Live Face"):
            # 注意要用 gr.components.Camera 或 gr.components…
            camera = gr.components.Camera(label="請對準鏡頭 (Live)")
            out_img = gr.Image(label="擷取畫面")
            out_lbl = gr.Label(label="檢測到的情緒")
            camera.change(
                fn=analyze_face,
                inputs=camera,
                outputs=[out_img, out_lbl],
                live=True
            )

        with gr.TabItem("🎤 上傳語音檔"):
            wav = gr.File(label="選擇 .wav 檔案", file_types=[".wav"])
            wav_btn = gr.Button("開始分析")
            wav_out = gr.Textbox(label="語音偵測到的情緒")
            wav_btn.click(fn=analyze_audio, inputs=wav, outputs=wav_out)

        with gr.TabItem("⌨️ 輸入文字"):
            txt = gr.Textbox(label="在此輸入文字", lines=3)
            txt_btn = gr.Button("開始分析")
            txt_out = gr.Textbox(label="文字偵測到的情緒")
            txt_btn.click(fn=analyze_text, inputs=txt, outputs=txt_out)

# 啟動
if __name__ == "__main__":
    # Hugging Face Spaces 上不需要傳 host/port，直接 .launch() 即可
    demo.launch()