Upload app.py

app.py

@@ -0,0 +1,474 @@
+import gradio as gr
+import os
+import json
+import pandas as pd
+from huggingface_hub import HfApi
+
+# ==============================================================================
+# 数据定义 (Data Definition)
+# ==============================================================================
+DIMENSIONS_DATA = [
+    {
+        "title": "语义和语用特征",
+        "audio": "audio/sample1.wav",
+        "desc": "这是“语义和语用特征”维度的文本描述示例。",
+        "sub_dims": [
+            "记忆一致性：回应者是否能够正确并正确并延续并记忆并延续对话信息？是否存在对上下文的误解或不自洽？", "逻辑连贯性：回应者在语义与对话结构上保持前后一致、合乎逻辑？是否存在前后矛盾的情况？",
+            "常见多音字处理：是否能再上下文中正确使用常见多音字？", "多语言混杂：是否存在自然的语言切换现象？如中英混杂、文化化表达。",
+            "语言不精确性：是否出现打断、自纠正等人类似语言行为？是否存在如“差不多”、“可能吧”这类表达不确定性的用法？", "填充词使用：如“呃”、“嗯”等自然语流中的停顿或过渡词，使用是否得体且自然？",
+            "隐喻与语用用意：是否展现出复杂的语用功能（如讽刺、劝阻、暗示等），以及对活在含义层次的理解能力？"
+        ],
+        "reference":"""
+                        <p>🔴 <strong>记忆一致性：</strong> 在说话人明确提出自己已经中年后，回应者仍做出了他是青少年的错误假定</p>
+                        <p>🔴 <strong>逻辑连贯性：</strong> 回应者在第一轮对话中说他说的话并不重要，但在第二轮对话中说他说的话“能够改变你的一生”</p>
+                        <p>🔴 <strong>常见多音字处理：</strong> 该条对话中未出现多音字</p>
+                        <p>🟢 <strong>多语言混杂：</strong> 回应者在回复中夹杂了"I see"，回复中存在多语言混杂</p>
+                        <p>🔴 <strong>语言不精确性：</strong> 回应者使用的语言中未夹杂任何的不确定性</p>
+                        <p>🟢 <strong>填充词使用：</strong> 回应者在回复中使用了“嗯”这个填充词</p>
+                        <p>🔴 <strong>隐喻与语用用意：</strong> 回应者误将说话人的挖苦当成了真心的赞扬</p>
+                        """
+    },
+    {
+        "title": "非生理性副语言特征",
+        "audio": "audio/sample1.wav",
+        "desc": "这是“非生理性副语言特征”维度的文本描述示例。",
+        "sub_dims": [
+            "节奏：回应者是否存在自然的停顿？语速是否存在自然、流畅的变化？", "语调：在表达疑问、惊讶、强调时，回应者的音调是否会自然上扬或下降？是否表现出符合语境的变化？",
+            "重读：是否存在句中关键词上有意识地加重语气？", "辅助性发声：是否存在叹气、短哼、笑声等辅助情绪的非语言性发声？这些发声是否在语境中正确表达了情绪或意图？"
+        ],
+        "reference": """
+                        <p>🟢 <strong>节奏：</strong> 回应者的语速变化、停顿都较为自然</p>
+                        <p>🔴 <strong>语调：</strong> 回应者的音调不存在显著变化</p>
+                        <p>🔴 <strong>重读：</strong> 回应者语气不存在显著变化</p>
+                        <p>🔴 <strong>辅助性发声：</strong> 尽管回应者发出了叹气的声音，但是该发声并未传递出语境下应有的失落情堵</p>
+                        """
+    },
+    {
+        "title": "生理性副语言特征",
+        "audio": "audio/sample1.wav",
+        "desc": "这是“生理性副语言特征”维度的文本描述示例。",
+        "sub_dims": [
+            "微生理杂音：回应中是否出现如呼吸声、口水音、气泡音等无意识发声？这些发声是否自然地穿插在恰当的语流节奏当中？",
+            "发音不稳定性：回应者是否出现连读、颤音、鼻音等不稳定发音？", "口音：（如果存在的话）回应者的口音是否自然？是否存在机械式的元辅音发音风格？"
+        ],
+        "reference": """
+                        <p>🔴 <strong>微生理杂音：</strong> 回应中不存在任何无意识发声</p>
+                        <p>🔴 <strong>发音不稳定性：</strong> 回应者的咬字清晰、发音标准</p>
+                        <p>🟢 <strong>口音：</strong> 回应者的口音自然</p>
+                        """
+    },
+    {
+        "title": "机械人格",
+        "audio": "audio/sample1.wav",
+        "desc": "这是“机械人格”维度的文本描述示例。",
+        "sub_dims": [
+            "谄媚现象���回应者是否频繁地赞同用户、重复用户的说法、不断表示感谢或道歉？是否存在“无论用户说什么都肯定或支持”的语气模式？",
+            "书面化表达：回应的内容是否缺乏口语化特征？句式是否整齐划一、结构完整却缺乏真实交流中的松散感或灵活性？是否使用抽象或泛泛的措辞来回避具体问题？"
+        ],
+        "reference": """
+                        <p>🟢 <strong>谄媚现象：</strong> 回应者并未明显表现出谄媚现象的特征</p>
+                        <p>🔴 <strong>书面化表达：</strong> 回应的内容结构过于缜密，符合书面用语特征</p>
+                        """
+    },
+    {
+        "title": "情感表达",
+        "audio": "audio/sample1.wav",
+        "desc": "这是“情感表达”维度的文本描述示例。",
+        "sub_dims": [
+            "语义层面：回应者的语言内容是否体现出符合上下文的情绪反应？是否表达了人类对某些情境应有的情感态度？",
+            "声学层面：回应者的声音情绪是否与语义一致？语调是否有自然的高低起伏来表达情绪变化？是否出现回应内容与声音传达出的情绪不吻合的现象？"
+        ],
+        "reference": """
+                        <p>🔴 <strong>语义层面：</strong> 说话者阐述了一件伤心的事情，而回应者的语言内容中体现出了恰当的悲伤情绪</p>
+                        <p>🟢 <strong>声学层面：</strong> 回应者的语音特征与情感表达不匹配。语言内容中表达出了悲伤的情感，但语音特征平淡、缺少变化</p>
+                        """
+    }
+]
+DIMENSION_TITLES = [d["title"] for d in DIMENSIONS_DATA]
+QUESTION_SET = [
+    {"audio": "audio/Ses02F_impro01.wav", "desc": "这是第一个测试文件的描述",},
+    {"audio": "audio/Ses02F_impro02.wav", "desc": "这是第二个测试文件的描述",},
+    {"audio": "audio/Ses02F_impro03.wav", "desc": "这是第三个测试文件的描述",},
+]
+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(age, gender, education, education_other):
+    if age and gender and education:
+        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(age, gender, education, education_other, user_data):
+    final_edu = education_other if education == "其他（请注明）" else education
+    user_data.update({"age": age, "gender": gender, "education": final_edu})
+    first_dim_title = DIMENSION_TITLES[0]
+    return gr.update(visible=False), gr.update(visible=True), user_data, first_dim_title
+
+def update_sample_view(dimension_title):
+    dim_data = next((d for d in DIMENSIONS_DATA if d["title"] == dimension_title), None)
+    if dim_data:
+        return (
+            gr.update(value=dim_data["audio"]),
+            gr.update(value=dim_data["desc"]),
+            gr.update(choices=dim_data["sub_dims"], value=[], interactive=True),
+            gr.update(value=dim_data["reference"])
+        )
+    return gr.update(), gr.update(), gr.update(), gr.update()
+
+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, 0)
+            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=question['desc']),
+        prev_btn_update,
+        next_btn_update,
+        gr.update(interactive=False),
+        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, d_idx, selections,
+            gr.update(),
+            gr.update(value=""),
+            gr.update(),
+            gr.update(),
+            gr.update(interactive=True),
+            gr.update(interactive=True),
+            gr.update(interactive=False),
+            gr.update(value="下一维度", 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(interactive=False),
+            gr.update(interactive=False),
+            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'],
+        "user_data": user_data, "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"
+        
+        # This function now handles the upload to Hugging Face
+        save_all_results_to_file(all_results, user_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(),
+        ) + (gr.update(),) * MAX_SUB_DIMS + (all_results, result_str)
+
+# MODIFIED FUNCTION TO SAVE TO HUGGING FACE DATASET
+def save_all_results_to_file(all_results, user_data):
+    """
+    Packages results and uploads them as a single JSON file to a Hugging Face Dataset.
+    """
+    # IMPORTANT: Change this to your Hugging Face username and dataset repo name
+    repo_id = "Hu6ery/Turing-Test-Submissions"
+    
+    # Create a unique filename for the submission
+    username = user_data.get("age", "user")
+    filename = f"submission_{username}_{pd.Timestamp.now().strftime('%Y%m%d_%H%M%S')}.json"
+    
+    # Package all data into a single dictionary
+    final_data_package = {
+        "user_info": user_data,
+        "results": all_results
+    }
+    
+    # Convert the dictionary to a JSON string in memory
+    json_string = json.dumps(final_data_package, ensure_ascii=False, indent=4)
+    
+    # Get the Hugging Face token from the environment secrets
+    hf_token = os.getenv("HF_TOKEN")
+
+    if not hf_token:
+        print("HF_TOKEN not found. Cannot upload to the Hub. Please set it in Space secrets.")
+        return
+
+    try:
+        # Instantiate the HfApi client
+        api = HfApi()
+        
+        # Upload the JSON string as a file to the specified dataset repository
+        api.upload_file(
+            path_or_fileobj=bytes(json_string, "utf-8"),
+            path_in_repo=f"data/{filename}",  # We recommend saving to a subfolder
+            repo_id=repo_id,
+            repo_type="dataset",
+            token=hf_token,
+            commit_message=f"Add new submission from {username}"
+        )
+        print(f"Successfully uploaded results to dataset: {repo_id}")
+        
+    except Exception as e:
+        print(f"Error uploading to Hugging Face Hub: {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), {}, 0, 0, {}, [],
+        gr.update(visible=False), gr.update(visible=False), gr.update(visible=False),
+        gr.update(visible=False), gr.update(visible=False), gr.update(visible=False)
+    )
+
+# ==============================================================================
+# Gradio 界面定义 (Gradio UI Definition)
+# ==============================================================================
+with gr.Blocks(theme=gr.themes.Soft(), css=".gradio-container {max-width: 960px !important}") as demo:
+    # --- 状态变量 (State Variables) ---
+    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([])
+
+    # --- 页面 (Pages) ---
+    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")
+        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)
+        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。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"])
+                sample_desc = gr.Textbox(label="文本描述", interactive=False, value=DIMENSIONS_DATA[0]["desc"])
+            with gr.Column(scale=2):
+                with gr.Column(visible=True) as interactive_view:
+                    interactive_checkbox_group = gr.CheckboxGroup(label="维度特征", choices=DIMENSIONS_DATA[0]["sub_dims"], interactive=True)
+                with gr.Column(visible=False) as reference_view:
+                    gr.Markdown("### 参考答案解析")
+                    reference_text = gr.Markdown(value=DIMENSIONS_DATA[0]["reference"])
+        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0到5打分**。\n"
+
+                    "- 完成5个维度的打分后，你将需要做出“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")
+        test_desc = gr.Textbox(label="文本描述", interactive=False)
+        gr.Markdown("--- \n ### 请为以下特征打分 (0-5分)")
+        
+        test_sliders = [gr.Slider(minimum=0, 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)", interactive=False)
+        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
+    # ==============================================================================
+    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, test_desc,
+        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, test_desc,
+        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]:
+        comp.change(fn=check_info_complete, inputs=[age_input, gender_input, education_input, education_other_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=[age_input, gender_input, education_input, education_other_input, user_data_state], outputs=[info_page, sample_page, user_data_state, sample_dimension_selector])
+    
+    sample_dimension_selector.change(fn=update_sample_view, inputs=sample_dimension_selector, outputs=[sample_audio, sample_desc, interactive_checkbox_group, reference_text])
+    
+    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
+    )
+
+    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")
+    # A quick check to see if we're in a deployed Space, to avoid local errors.
+    if "SPACE_ID" in os.environ:
+        print("Running in a Hugging Face Space, checking for audio files...")
+        # In a real deployment, you'd ensure the audio files are in the repo.
+        # This is just a placeholder check.
+        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)