Update app.py

app.py

@@ -89,6 +89,715 @@ DIMENSIONS_DATA = [
 
 DIMENSION_TITLES = [d["title"] for d in DIMENSIONS_DATA]
 
+def load_or_initialize_count_json(audio_paths):
+    if os.path.exists(COUNT_JSON_PATH):
+        with open(COUNT_JSON_PATH, "r", encoding="utf-8") as f:
+            # 使用 object_pairs_hook 保持原始顺序
+            count_data = json.load(f, object_pairs_hook=collections.OrderedDict)
+    else:
+        count_data = collections.OrderedDict()
+
+    updated = False
+    for path in audio_paths:
+        filename = os.path.basename(path)
+        if filename not in count_data:
+            count_data[filename] = 0
+            updated = True
+
+    if updated or not os.path.exists(COUNT_JSON_PATH):
+        with open(COUNT_JSON_PATH, "w", encoding="utf-8") as f:
+            # 确保写入时也保持顺序
+            json.dump(count_data, f, indent=4, ensure_ascii=False)
+    
+    return count_data
+
+def append_cache_buster(audio_path):
+    return f"{audio_path}?t={int(time.time() * 1000)}"
+
+def sample_audio_paths(audio_paths, count_data, k=5, max_count=1):
+    eligible_paths = [p for p in audio_paths if count_data.get(os.path.basename(p), 0) < max_count]
+    
+    if len(eligible_paths) < k:
+        raise ValueError(f"可用音频数量不足（只剩 {len(eligible_paths)} 条 count<{max_count} 的音频），无法抽取 {k} 条")
+
+    eligible_paths_copy = eligible_paths.copy()
+    
+    random.seed(int(time.time()))
+    
+    selected = random.sample(eligible_paths_copy, k)
+
+    for path in selected:
+        filename = os.path.basename(path)
+        count_data[filename] = count_data.get(filename, 0) + 1
+
+    with open(COUNT_JSON_PATH, "w", encoding="utf-8") as f:
+        json.dump(count_data, f, indent=4, ensure_ascii=False)
+
+    return selected, count_data
+
+count_data = load_or_initialize_count_json(all_data_audio_paths)
+selected_audio_paths, updated_count_data = sample_audio_paths(all_data_audio_paths, count_data, k=5)
+
+
+QUESTION_SET = [
+    {"audio": path, "desc": f"这是音频文件 {os.path.basename(path)} 的描述"} 
+    for path in selected_audio_paths
+]
+
+MAX_SUB_DIMS = max(len(d['sub_dims']) for d in DIMENSIONS_DATA)
+
+# ==============================================================================
+# 功能函数定义 (Function Definitions)
+# ==============================================================================
+def start_challenge():
+    return gr.update(visible=False), gr.update(visible=True)
+
+def toggle_education_other(choice):
+    is_other = (choice == "其他（请注明）")
+    return gr.update(visible=is_other, interactive=is_other, value="")
+
+def check_info_complete(username, age, gender, education, education_other, ai_experience):
+    if username.strip() and age and gender and education and ai_experience:
+        if education == "其他（请注明）" and not education_other.strip():
+            return gr.update(interactive=False)
+        return gr.update(interactive=True)
+    return gr.update(interactive=False)
+
+def show_sample_page_and_init(username, age, gender, education, education_other, ai_experience, user_data):
+    final_edu = education_other if education == "其他（请注明）" else education
+    user_data.update({
+        "username": username.strip(),
+        "age": age, 
+        "gender": gender, 
+        "education": final_edu,
+        "ai_experience": ai_experience
+    })
+    first_dim_title = DIMENSION_TITLES[0]
+    
+    initial_updates = update_sample_view(first_dim_title)
+    
+    return [
+        gr.update(visible=False), gr.update(visible=True), user_data, first_dim_title
+    ] + initial_updates
+
+def update_sample_view(dimension_title):
+    dim_data = next((d for d in DIMENSIONS_DATA if d["title"] == dimension_title), None)
+    if dim_data:
+        audio_up = gr.update(value=dim_data["audio"])
+        # audio_up = gr.update(value=append_cache_buster(dim_data["audio"]))
+        interactive_view_up = gr.update(visible=True)
+        reference_view_up = gr.update(visible=False)
+        reference_btn_up = gr.update(value="参考")
+        sample_slider_ups = []
+        ref_slider_ups = []
+        scores = dim_data.get("reference_scores", [])
+
+        for i in range(MAX_SUB_DIMS):
+            if i < len(dim_data['sub_dims']):
+                label = dim_data['sub_dims'][i]
+                score = scores[i] if i < len(scores) else 0
+                sample_slider_ups.append(gr.update(visible=True, label=label, value=3))
+                ref_slider_ups.append(gr.update(visible=True, label=label, value=score))
+            else:
+                sample_slider_ups.append(gr.update(visible=False, value=0))
+                ref_slider_ups.append(gr.update(visible=False, value=0))
+        return [audio_up, interactive_view_up, reference_view_up, reference_btn_up] + sample_slider_ups + ref_slider_ups
+    empty_updates = [gr.update()] * 4
+    slider_empty_updates = [gr.update()] * (MAX_SUB_DIMS * 2)
+    return empty_updates + slider_empty_updates
+
+def update_test_dimension_view(d_idx, selections):
+    dimension = DIMENSIONS_DATA[d_idx]
+    progress_d = f"维度 {d_idx + 1} / {len(DIMENSIONS_DATA)}: **{dimension['title']}**"
+    
+    existing_scores = selections.get(dimension['title'], {})
+    
+    slider_updates = []
+    for i in range(MAX_SUB_DIMS):
+        if i < len(dimension['sub_dims']):
+            sub_dim_label = dimension['sub_dims'][i]
+            value = existing_scores.get(sub_dim_label, 3)
+            slider_updates.append(gr.update(visible=True, label=sub_dim_label, value=value))
+        else:
+            slider_updates.append(gr.update(visible=False, value=0))
+            
+    prev_btn_update = gr.update(interactive=(d_idx > 0))
+    next_btn_update = gr.update(
+        value="进入最终判断" if d_idx == len(DIMENSIONS_DATA) - 1 else "下一维度",
+        interactive=True
+    )
+    
+    return [gr.update(value=progress_d), prev_btn_update, next_btn_update] + slider_updates
+
+def init_test_question(user_data, q_idx):
+    d_idx = 0
+    question = QUESTION_SET[q_idx]
+    progress_q = f"第 {q_idx + 1} / {len(QUESTION_SET)} 题"
+    
+    initial_updates = update_test_dimension_view(d_idx, {})
+    dim_title_update, prev_btn_update, next_btn_update = initial_updates[:3]
+    slider_updates = initial_updates[3:]
+    
+    return (
+        gr.update(visible=False),
+        gr.update(visible=True),
+        gr.update(visible=False),
+        gr.update(visible=False),
+        q_idx, d_idx, {},
+        gr.update(value=progress_q),
+        dim_title_update,
+        gr.update(value=question['audio']),
+        # gr.update(value=append_cache_buster(question['audio'])),
+        prev_btn_update,
+        next_btn_update,
+        gr.update(value=None), # BUG FIX: Changed from "" to None to correctly clear the radio button
+        gr.update(interactive=False),
+    ) + tuple(slider_updates)
+
+def navigate_dimensions(direction, q_idx, d_idx, selections, *slider_values):
+    current_dim_data = DIMENSIONS_DATA[d_idx]
+    current_sub_dims = current_dim_data['sub_dims']
+    scores = {sub_dim: slider_values[i] for i, sub_dim in enumerate(current_sub_dims)}
+    selections[current_dim_data['title']] = scores
+
+    new_d_idx = d_idx + (1 if direction == "next" else -1)
+
+    if direction == "next" and d_idx == len(DIMENSIONS_DATA) - 1:
+        return (
+            gr.update(visible=False),
+            gr.update(visible=True),
+            q_idx, new_d_idx, selections,
+            gr.update(),
+            gr.update(),
+            gr.update(),
+            gr.update(interactive=True),
+            gr.update(interactive=False),
+            gr.update(interactive=False),
+            gr.update(interactive=False),
+        ) + (gr.update(),) * MAX_SUB_DIMS
+
+    else:
+        view_updates = update_test_dimension_view(new_d_idx, selections)
+        dim_title_update, prev_btn_update, next_btn_update = view_updates[:3]
+        slider_updates = view_updates[3:]
+
+        return (
+            gr.update(), gr.update(),
+            q_idx, new_d_idx, selections,
+            gr.update(),
+            dim_title_update,
+            gr.update(),
+            gr.update(),
+            gr.update(),
+            prev_btn_update,
+            next_btn_update,
+        ) + tuple(slider_updates)
+
+def submit_question_and_advance(q_idx, d_idx, selections, final_choice, all_results, user_data):
+    selections["final_choice"] = final_choice
+    
+    final_question_result = {
+        "question_id": q_idx, "audio_file": QUESTION_SET[q_idx]['audio'],
+        "selections": selections
+    }
+    all_results.append(final_question_result)
+    
+    q_idx += 1
+    
+    if q_idx < len(QUESTION_SET):
+        init_q_updates = init_test_question(user_data, q_idx)
+        return init_q_updates + (all_results, gr.update(value=""))
+    else:
+        result_str = "### 测试全部完成！\n\n你的提交结果概览：\n"
+        for res in all_results:
+            # result_str += f"\n#### 题目: {res['audio_file']}\n"
+            result_str += f"##### 最终判断: **{res['selections'].get('final_choice', '未选择')}**\n"
+            for dim_title, dim_data in res['selections'].items():
+                if dim_title == 'final_choice': continue
+                result_str += f"- **{dim_title}**:\n"
+                for sub_dim, score in dim_data.items():
+                    result_str += f"  - *{sub_dim[:20]}...*: {score}/5\n"
+        
+        # save_all_results_to_file(all_results, user_data)
+        save_all_results_to_file(all_results, user_data, count_data=updated_count_data)
+        
+        return (
+            gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=True),
+            q_idx, d_idx, {},
+            gr.update(), gr.update(), gr.update(), gr.update(), gr.update(),
+            gr.update(), gr.update(),
+        ) + (gr.update(),) * MAX_SUB_DIMS + (all_results, result_str)
+
+def save_all_results_to_file(all_results, user_data, count_data=None):
+    repo_id = "intersteller2887/Turing-test-dataset"
+    username = user_data.get("username", "user")
+    timestamp = pd.Timestamp.now().strftime('%Y%m%d_%H%M%S')
+    submission_filename = f"submissions_{username}_{timestamp}.json"
+    
+    final_data_package = {
+        "user_info": user_data,
+        "results": all_results
+    }
+    json_string = json.dumps(final_data_package, ensure_ascii=False, indent=4)
+    hf_token = os.getenv("HF_TOKEN")
+
+    if not hf_token:
+        print("HF_TOKEN not found. Cannot upload to the Hub.")
+        return
+
+    try:
+        api = HfApi()
+
+        # 上传 submission 文件
+        api.upload_file(
+            path_or_fileobj=bytes(json_string, "utf-8"),
+            path_in_repo=f"submissions/{submission_filename}",
+            repo_id=repo_id,
+            repo_type="dataset",
+            token=hf_token,
+            commit_message=f"Add new submission from {username}"
+        )
+        print(f"上传成功: {submission_filename}")
+
+        # 上传 count.json（如果提供）
+        if count_data:
+            with open(COUNT_JSON_PATH, "w", encoding="utf-8") as f:
+                json.dump(count_data, f, indent=4, ensure_ascii=False)
+
+            api.upload_file(
+                path_or_fileobj=COUNT_JSON_PATH,
+                path_in_repo=COUNT_JSON_REPO_PATH,
+                repo_id=repo_id,
+                repo_type="dataset",
+                token=hf_token,
+                commit_message=f"Update count.json after submission by {username}"
+            )
+            print("count.json 上传成功")
+
+    except Exception as e:
+        print(f"上传出错: {e}")
+
+
+def toggle_reference_view(current):
+    if current == "参考":
+        return gr.update(visible=False), gr.update(visible=True), gr.update(value="返回")
+    else:
+        return gr.update(visible=True), gr.update(visible=False), gr.update(value="参考")
+
+def back_to_welcome():
+    return (
+        gr.update(visible=True),   # welcome_page
+        gr.update(visible=False),  # info_page
+        gr.update(visible=False),  # sample_page
+        gr.update(visible=False),  # pretest_page
+        gr.update(visible=False),  # test_page
+        gr.update(visible=False),  # final_judgment_page
+        gr.update(visible=False),  # result_page
+        {},                        # user_data_state
+        0,                         # current_question_index
+        0,                         # current_test_dimension_index
+        {},                        # current_question_selections
+        []                         # test_results
+    )
+
+# ==============================================================================
+# Gradio 界面定义 (Gradio UI Definition)
+# ==============================================================================
+with gr.Blocks(theme=gr.themes.Soft(), css=".gradio-container {max-width: 960px !important}") as demo:
+    user_data_state = gr.State({})
+    current_question_index = gr.State(0)
+    current_test_dimension_index = gr.State(0)
+    current_question_selections = gr.State({})
+    test_results = gr.State([])
+
+    welcome_page = gr.Column(visible=True)
+    info_page = gr.Column(visible=False)
+    sample_page = gr.Column(visible=False)
+    pretest_page = gr.Column(visible=False)
+    test_page = gr.Column(visible=False)
+    final_judgment_page = gr.Column(visible=False)
+    result_page = gr.Column(visible=False)
+    pages = {
+        "welcome": welcome_page, "info": info_page, "sample": sample_page,
+        "pretest": pretest_page, "test": test_page, "final_judgment": final_judgment_page,
+        "result": result_page
+    }
+
+    with welcome_page:
+        gr.Markdown("# AI 识破者\n你将听到一系列对话，请判断哪个回应者是 AI。")
+        start_btn = gr.Button("开始挑战", variant="primary")
+
+    with info_page:
+        gr.Markdown("## 请提供一些基本信息")
+        username_input = gr.Textbox(label="用户名", placeholder="请输入一个昵称或代号")
+        age_input = gr.Radio(["18岁以下", "18-25岁", "26-35岁", "36-50岁", "50岁以上"], label="年龄")
+        gender_input = gr.Radio(["男", "女", "其他"], label="性别")
+        education_input = gr.Radio(["高中及以下", "本科", "硕士", "博士", "其他（请注明）"], label="学历")
+        education_other_input = gr.Textbox(label="请填写你的学历", visible=False, interactive=False)
+        ai_experience_input = gr.Radio(["从未使用过", "偶尔接触（如看别人用）", "使用过几次，了解基本功能", "经常使用，有一定操作经验", "非常熟悉，深入使用过多个 AI 工具"], label="对 AI 工具的熟悉程度")
+        submit_info_btn = gr.Button("提交并开始学习样例", variant="primary", interactive=False)
+
+    with sample_page:
+
+        gr.Markdown("## 样例分析\n请选择一个维度进行学习和打分练习。所有维度共用同一个样例音频。")
+        sample_dimension_selector = gr.Radio(DIMENSION_TITLES, label="选择学习维度", value=DIMENSION_TITLES[0])
+        with gr.Row():
+            with gr.Column(scale=1):
+                sample_audio = gr.Audio(label="样例音频", value=DIMENSIONS_DATA[0]["audio"])
+            with gr.Column(scale=2):
+                with gr.Column(visible=True) as interactive_view:
+                    gr.Markdown("#### 请为以下特征打分 (1-5分。1对应机器，5对应人类)")
+                    sample_sliders = [gr.Slider(minimum=1, maximum=5, step=1, label=f"Sub-dim {i+1}", visible=False, interactive=True) for i in range(MAX_SUB_DIMS)]
+                with gr.Column(visible=False) as reference_view:
+                    gr.Markdown("### 参考答案解析 (1-5分。1对应机器，5对应人类)")
+                    reference_sliders = [gr.Slider(minimum=1, maximum=5, step=1, label=f"Sub-dim {i+1}", visible=False, interactive=False) for i in range(MAX_SUB_DIMS)]
+        with gr.Row():
+            reference_btn = gr.Button("参考")
+            go_to_pretest_btn = gr.Button("我明白了，开始测试", variant="primary")
+
+    with pretest_page:
+        gr.Markdown("## 测试说明\n"
+                      "- 对于每一道题，你都需要对全部 **5 个维度** 进行评估。\n"
+                      "- 在每个维度下，请为出现的每个特征 **从1到5打分。\n"
+                      "- **评分解释如下：**\n"
+                      "  - **1 分：极度符合机器特征**；\n"
+                      "  - **2 分：较为符合机器特征**；\n"
+                      "  - **3 分：无明显人类或机器倾向或特征无体现**；\n"
+                      "  - **4 分：较为符合人类特征**；\n"
+                      "  - **5 分：极度符合人类特征**。\n"
+                      "- 完成所有维度后，请根据整体印象对回应方的身份做出做出“人类”或“机器人”的 **最终判断**。\n"
+                      "- 你可以使用“上一维度”和“下一维度”按钮在5个维度间自由切换和修改分数。")
+        go_to_test_btn = gr.Button("开始测试", variant="primary")
+
+    with test_page:
+        gr.Markdown("## 正式测试")
+        question_progress_text = gr.Markdown()
+        test_dimension_title = gr.Markdown()
+        test_audio = gr.Audio(label="测试音频")
+        gr.Markdown("--- \n ### 请为以下特征打分 (1-5分。1对应机器，5对应人类)")
+        test_sliders = [gr.Slider(minimum=1, maximum=5, step=1, label=f"Sub-dim {i+1}", visible=False, interactive=True) for i in range(MAX_SUB_DIMS)]
+        with gr.Row():
+            prev_dim_btn = gr.Button("上一维度")
+            next_dim_btn = gr.Button("下一维度", variant="primary")
+
+    with final_judgment_page:
+        gr.Markdown("## 最终判断")
+        gr.Markdown("您已完成对所有维度的评分。请根据您的综合印象，做出最终判断。")
+        final_human_robot_radio = gr.Radio(["👤 人类", "🤖 机器人"], label="请判断回应者类型 (必填)")
+        submit_final_answer_btn = gr.Button("提交本题答案", variant="primary", interactive=False)
+
+    with result_page:
+        gr.Markdown("## 测试完成")
+        result_text = gr.Markdown()
+        back_to_welcome_btn = gr.Button("返回主界面", variant="primary")
+
+    # ==============================================================================
+    # 事件绑定 (Event Binding) & IO 列表定义
+    # ==============================================================================
+    sample_init_outputs = [
+        info_page, sample_page, user_data_state, sample_dimension_selector,
+        sample_audio, interactive_view, reference_view, reference_btn
+    ] + sample_sliders + reference_sliders
+    
+    test_init_outputs = [
+        pretest_page, test_page, final_judgment_page, result_page,
+        current_question_index, current_test_dimension_index, current_question_selections,
+        question_progress_text, test_dimension_title, test_audio,
+        prev_dim_btn, next_dim_btn,
+        final_human_robot_radio, submit_final_answer_btn,
+    ] + test_sliders
+
+    nav_inputs = [current_question_index, current_test_dimension_index, current_question_selections] + test_sliders
+    nav_outputs = [
+        test_page, final_judgment_page,
+        current_question_index, current_test_dimension_index, current_question_selections,
+        question_progress_text, test_dimension_title, test_audio,
+        final_human_robot_radio, submit_final_answer_btn,
+        prev_dim_btn, next_dim_btn,
+    ] + test_sliders
+    
+    full_outputs_with_results = test_init_outputs + [test_results, result_text]
+
+    start_btn.click(fn=start_challenge, outputs=[welcome_page, info_page])
+    
+    for comp in [age_input, gender_input, education_input, education_other_input, ai_experience_input]:
+        comp.change(
+            fn=check_info_complete,
+            inputs=[username_input, age_input, gender_input, education_input, education_other_input, ai_experience_input],
+            outputs=submit_info_btn
+        )
+    
+    education_input.change(fn=toggle_education_other, inputs=education_input, outputs=education_other_input)
+    
+    submit_info_btn.click(
+        fn=show_sample_page_and_init,
+        inputs=[username_input, age_input, gender_input, education_input, education_other_input, ai_experience_input, user_data_state],
+        outputs=sample_init_outputs
+    )
+    
+    sample_dimension_selector.change(
+        fn=update_sample_view, 
+        inputs=sample_dimension_selector, 
+        outputs=[sample_audio, interactive_view, reference_view, reference_btn] + sample_sliders + reference_sliders
+    )
+
+    reference_btn.click(
+        fn=toggle_reference_view, 
+        inputs=reference_btn, 
+        outputs=[interactive_view, reference_view, reference_btn]
+    )
+    
+    go_to_pretest_btn.click(lambda: (gr.update(visible=False), gr.update(visible=True)), outputs=[sample_page, pretest_page])
+    
+    go_to_test_btn.click(
+        fn=lambda user: init_test_question(user, 0) + ([], gr.update()),
+        inputs=[user_data_state],
+        outputs=full_outputs_with_results
+    )
+    
+    prev_dim_btn.click(
+        fn=lambda q,d,s, *sliders: navigate_dimensions("prev", q,d,s, *sliders),
+        inputs=nav_inputs, outputs=nav_outputs
+    )
+    
+    next_dim_btn.click(
+        fn=lambda q,d,s, *sliders: navigate_dimensions("next", q,d,s, *sliders),
+        inputs=nav_inputs, outputs=nav_outputs
+    )
+
+    final_human_robot_radio.change(
+        fn=lambda choice: gr.update(interactive=bool(choice)),
+        inputs=final_human_robot_radio,
+        outputs=submit_final_answer_btn
+    )
+
+    submit_final_answer_btn.click(
+        fn=submit_question_and_advance,
+        inputs=[current_question_index, current_test_dimension_index, current_question_selections, final_human_robot_radio, test_results, user_data_state],
+        outputs=full_outputs_with_results
+    )
+
+    back_to_welcome_btn.click(fn=back_to_welcome, outputs=list(pages.values()) + [user_data_state, current_question_index, current_test_dimension_index, current_question_selections, test_results])
+
+# ==============================================================================
+# 程序入口 (Entry Point)
+# ==============================================================================
+if __name__ == "__main__":
+    if not os.path.exists("audio"):
+        os.makedirs("audio")
+    if "SPACE_ID" in os.environ:
+        print("Running in a Hugging Face Space, checking for audio files...")
+        all_files = [q["audio"] for q in QUESTION_SET] + [d["audio"] for d in DIMENSIONS_DATA]
+        for audio_file in set(all_files):
+            if not os.path.exists(audio_file):
+                print(f"⚠️ Warning: Audio file not found: {audio_file}")
+    
+    demo.launch(debug=True)
+我还需要添加两个功能：第一，sample的音频文件直接计数为max（防止被抽取）
+第二，在最后点击提交按钮时，需要终止可能正在播放的音频
+我将帮你添加这两个功能：
+
+首先修改 load_or_initialize_count_json 函数，将样例音频文件直接设置为最大值：
+
+python
+def load_or_initialize_count_json(audio_paths):
+    if os.path.exists(COUNT_JSON_PATH):
+        with open(COUNT_JSON_PATH, "r", encoding="utf-8") as f:
+            # 使用 object_pairs_hook 保持原始顺序
+            count_data = json.load(f, object_pairs_hook=collections.OrderedDict)
+    else:
+        count_data = collections.OrderedDict()
+
+    updated = False
+    
+    # 获取所有样例音频文件名
+    sample_audio_files = {os.path.basename(d["audio"]) for d in DIMENSIONS_DATA}
+    
+    for path in audio_paths:
+        filename = os.path.basename(path)
+        if filename not in count_data:
+            # 如果是样例音频，直接设置为最大值
+            if filename in sample_audio_files:
+                count_data[filename] = 999  # 设置为很大的值，确保不会被选中
+            else:
+                count_data[filename] = 0
+            updated = True
+
+    if updated or not os.path.exists(COUNT_JSON_PATH):
+        with open(COUNT_JSON_PATH, "w", encoding="utf-8") as f:
+            # 确保写入时也保持顺序
+            json.dump(count_data, f, indent=4, ensure_ascii=False)
+    
+    return count_data
+添加音频停止功能，修改 Gradio 界面部分：
+
+在 submit_question_and_advance 函数中添加停止音频的逻辑：
+
+python
+def submit_question_and_advance(q_idx, d_idx, selections, final_choice, all_results, user_data):
+    selections["final_choice"] = final_choice
+    
+    final_question_result = {
+        "question_id": q_idx, "audio_file": QUESTION_SET[q_idx]['audio'],
+        "selections": selections
+    }
+    all_results.append(final_question_result)
+    
+    q_idx += 1
+    
+    if q_idx < len(QUESTION_SET):
+        init_q_updates = init_test_question(user_data, q_idx)
+        # 添加停止音频的逻辑
+        init_q_updates = list(init_q_updates)
+        init_q_updates[9] = None  # test_audio 的位置设为 None 来停止播放
+        return tuple(init_q_updates) + (all_results, gr.update(value=""))
+    else:
+        result_str = "### 测试全部完成！\n\n你的提交结果概览：\n"
+        for res in all_results:
+            result_str += f"##### 最终判断: **{res['selections'].get('final_choice', '未选择')}**\n"
+            for dim_title, dim_data in res['selections'].items():
+                if dim_title == 'final_choice': continue
+                result_str += f"- **{dim_title}**:\n"
+                for sub_dim, score in dim_data.items():
+                    result_str += f"  - *{sub_dim[:20]}...*: {score}/5\n"
+        
+        save_all_results_to_file(all_results, user_data, count_data=updated_count_data)
+        
+        return (
+            gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=True),
+            q_idx, d_idx, {},
+            gr.update(), gr.update(), None, gr.update(), gr.update(),  # 设置 test_audio 为 None 来停止播放
+            gr.update(), gr.update(),
+        ) + (gr.update(),) * MAX_SUB_DIMS + (all_results, result_str)
+在返回主界面时也停止音频：
+
+修改 back_to_welcome 函数：
+
+python
+def back_to_welcome():
+    return (
+        gr.update(visible=True),   # welcome_page
+        gr.update(visible=False),  # info_page
+        gr.update(visible=False),  # sample_page
+        gr.update(visible=False),  # pretest_page
+        gr.update(visible=False),  # test_page
+        gr.update(visible=False),  # final_judgment_page
+        gr.update(visible=False),  # result_page
+        {},                        # user_data_state
+        0,                         # current_question_index
+        0,                         # current_test_dimension_index
+        {},                        # current_question_selections
+        [],                        # test_results
+        None                       # 停止音频
+    )
+更新 back_to_welcome_btn 的点击事件：
+
+python
+back_to_welcome_btn.click(
+    fn=back_to_welcome, 
+    outputs=list(pages.values()) + [user_data_state, current_question_index, current_test_dimension_index, current_question_selections, test_results, test_audio]
+)
+这些修改将实现：
+
+样例音频文件不会被选为测试音频（因为它们的计数值被设为999）
+
+在提交答案或返回主界面时，当前播放的音频会被停止
+
+切换到新题目时，前一个音频会自动停止
+
+注意：Gradio 的音频组件设置为 None 时会停止当前播放。
+
+import gradio as gr
+import os
+import json
+import pandas as pd
+import random
+import shutil
+import time
+import collections
+from datasets import load_dataset
+from huggingface_hub import HfApi
+
+dataset = load_dataset("intersteller2887/Turing-test-dataset", split="train")
+
+target_audio_dir = "/home/user/app/audio"
+os.makedirs(target_audio_dir, exist_ok=True)
+COUNT_JSON_PATH = "/home/user/app/count.json"
+COUNT_JSON_REPO_PATH = "submissions/count.json"
+
+local_audio_paths = []
+
+for item in dataset:
+    src_path = item["audio"]["path"]
+    if src_path and os.path.exists(src_path):
+        filename = os.path.basename(src_path)
+        dst_path = os.path.join(target_audio_dir, filename)
+        if not os.path.exists(dst_path):
+            shutil.copy(src_path, dst_path)
+        local_audio_paths.append(dst_path)
+
+all_data_audio_paths = local_audio_paths
+
+sample1_audio_path = local_audio_paths[0]
+# sample1_audio_path = next((p for p in all_data_audio_paths if p.endswith("sample1.wav")), None)
+print(sample1_audio_path)
+
+# ==============================================================================
+# 数据定义 (Data Definition)
+# ==============================================================================
+
+DIMENSIONS_DATA = [
+    {
+        "title": "语义和语用特征",
+        "audio": sample1_audio_path,
+        "sub_dims": [
+            "记忆一致性：人类会选择性记忆并自我修正错误；机器出现前后矛盾时无法自主察觉或修正（如：遗忘关键细节但坚持错误答案）", "逻辑连贯性：人类逻辑自然流畅，允许合理跳跃；机器逻辑转折生硬或自相矛盾（如：突然切换话题无过渡）",
+            "读音正确性：人类大部分情况下发音正确、自然，会结合语境使用、区分多音字；机器存在不自然的发音错误，且对多音字语境的判断能力有限", "多语言混杂：人类多语言混杂流畅，且带有语境色彩；机器多语言混杂生硬，无语言切换逻辑",
+            "语言不精确性：人类说话时会使用带有犹豫语气的表达，且会出现自我修正的行为；机器的回应通常不存在模糊表达，回答准确、肯定", "填充词使用：人类填充词（如‘嗯’‘那个’）随机且带有思考痕迹；机器填充词规律重复或完全缺失",
+            "隐喻与语用用意：人类会使用隐喻、反语、委婉来表达多重含义；机器表达直白，仅能字面理解或生硬使用修辞，缺乏语义多样性"
+        ],
+        "reference_scores": [5, 5, 3, 3, 5, 5, 3]
+    },
+    {
+        "title": "非生理性副语言特征",
+        "audio": sample1_audio_path,
+        "sub_dims": [
+            "节奏：人类语速随语义起伏，偶尔卡顿或犹豫；机器节奏均匀，几乎无停顿或停顿机械", "语调：人类在表达疑问、惊讶、强调时，音调会自然上扬或下降；机器语调单一或变化过于规律，不符合语境",
+            "重读：人类会有意识地加强重要词语，从而突出信息焦点；机器的词语强度一致性强，或出现强调部位异常", "辅助性发声：人类会发出符合语境的非语言声音，如笑声、叹气等；机器的辅助性发声语义错误，或完全无辅助性发声"
+        ],
+        "reference_scores": [4, 5, 4, 3]
+    },
+    {
+        "title": "生理性副语言特征",
+        "audio": sample1_audio_path,
+        "sub_dims": [
+            "微生理杂音：人类说话存在呼吸声、口水音、气泡音等无意识发声，且自然地穿插在语流节奏当中；机器没有微生理杂音、语音过于干净，或添加不自然杂音",
+            "发音不稳定性：人类存在个体化波动（如偶尔咬字不清、鼻音丰富等）；机器发音过于标准或统一，缺乏个性", "口音：人类存在自然的地区口音或语音特征；机器元音辅音机械拼接，或口音模式统一无差异"
+        ],
+        "reference_scores": [3, 3, 4]
+    },
+    {
+        "title": "机械人格",
+        "audio": sample1_audio_path,
+        "sub_dims": [
+            "谄媚现象：人类会根据语境判断是否同意，有时提出不同意见；机器频繁同意、感谢、道歉，过度认同对方观点，缺乏真实互动感",
+            "书面化表达：人类表达灵活，；机器回应句式工整、规范，内容过于书面化、用词泛泛"
+        ],
+        "reference_scores": [5, 5]
+    },
+    {
+        "title": "情感表达",
+        "audio": sample1_audio_path,
+        "sub_dims": [
+            "语义层面：人类能对悲伤、开心等语境有符合人类的情感反应；机器回应情绪淡漠，或情感词泛泛、脱离语境",
+            "声学层面：人类音调、音量随情绪动态变化；机器情感语调模式化，或与语境不符"
+        ],
+        "reference_scores": [3, 3]
+    }
+]
+
+DIMENSION_TITLES = [d["title"] for d in DIMENSIONS_DATA]
+
 def load_or_initialize_count_json(audio_paths):
     if os.path.exists(COUNT_JSON_PATH):
         with open(COUNT_JSON_PATH, "r", encoding="utf-8") as f: