import gradio as gr import os import openai import time from pathlib import Path openai.api_key = os.getenv("OPENAI_API_KEY") client = openai.OpenAI() LLM_MODEL = "gpt-4o" ASR_MODEL = "gpt-4o-transcribe" TTS_MODEL = "gpt-4o-mini-tts" TTS_VOICE = "nova" # --- 預設辯論主題 (台灣時事) --- CURRENT_TW_TOPICS = [ "台灣是否應提高核能發電比例以應對能源需求？", "面對高房價，興建社會住宅是最佳解方嗎？", "台灣是否應放寬代理孕母的限制？ (涉及生育權、倫理爭議、法律規範)", "少子化下，延後退休年齡或引進更多外籍移工哪個更急迫？", "台灣是否應積極推動雙語教育（中英文）作為主要教學語言？", "健保制度改革：提高保費、部分負擔，或有其他永續方案？", "台灣是否應對網路言論進行更嚴格的管制？", "台灣的死刑制度是否應廢除？", ] # --- Helper 函數：呼叫 OpenAI API --- def call_asr(audio_filepath): if not audio_filepath: return "" try: with open(audio_filepath, "rb") as audio_file: transcript = client.audio.transcriptions.create(model=ASR_MODEL, file=audio_file) return transcript.text except Exception as e: print(f"ASR Error (OpenAI): {e}") return f"[語音辨識失敗: {e}]" def call_llm(topic, user_stance, messages): ai_stance = "反方" if user_stance == "正方" else "正方" system_prompt = f"你正在參與一場關於「{topic}」的辯論。你扮演的是堅定的「{ai_stance}」。請根據對話歷史，針對使用者的最新論點，提出具有批判性、質疑性或反駁性的回應。保持簡潔有力，專注於論證，字數控制在150字以內。" openai_messages = [{"role": "system", "content": system_prompt}] for msg in messages: role = msg.get("role") if msg.get("role") in ["user", "assistant"] else "user" content = msg.get("content", "") if content: openai_messages.append({"role": role, "content": content}) try: response = client.chat.completions.create(model=LLM_MODEL, messages=openai_messages, max_tokens=250, temperature=0.7) ai_response = response.choices[0].message.content.strip() return ai_response except Exception as e: print(f"LLM Error (OpenAI): {e}") return f"[AI 回應生成失敗: {e}]" def call_tts(text): try: if not text or not isinstance(text, str) or text.startswith("["): print(f"Skipping TTS for invalid text: {text}") return None speech_file_path = Path(f"/tmp/speech_{int(time.time() * 1000)}.mp3") response = client.audio.speech.create(model=TTS_MODEL, voice=TTS_VOICE, input=text) response.stream_to_file(speech_file_path) return str(speech_file_path) except Exception as e: print(f"TTS Error (OpenAI): {e}") return None # --- Gradio 主函數 --- def debate_turn(topic_from_dropdown, custom_topic, user_stance, user_input_text, user_input_audio, history): final_topic = "" if custom_topic and custom_topic.strip(): final_topic = custom_topic.strip() print(f"Using custom topic: {final_topic}") elif topic_from_dropdown: final_topic = topic_from_dropdown print(f"Using dropdown topic: {final_topic}") else: history.append(("[錯誤：請選擇或輸入一個辯論主題]", None)) return history, None, "", "" user_text = "" processed_audio_path = None if user_input_audio: print(f"Processing audio input: {user_input_audio}") processed_audio_path = user_input_audio user_text = call_asr(processed_audio_path) history.append(((processed_audio_path,), None)) if user_text.startswith("["): history.append((user_text, None)) return history, None, "", final_topic if not user_text or user_text.startswith("["): if user_input_text: user_text = user_input_text if not processed_audio_path: history.append((user_text, None)) elif user_input_text: user_text = user_input_text history.append((f"(改用文字輸入: {user_text})", None)) else: if not processed_audio_path: history.append(("[錯誤：請提供文字或語音論點]", None)) return history, None, "", final_topic if not isinstance(user_text, str) or user_text.startswith("["): print("Invalid user text, stopping turn.") if history and isinstance(history[-1][0], tuple) and history[-1][1] is None: history.append((f"[無法處理用戶輸入: {user_text}]", None)) return history, None, "", final_topic llm_messages = [] for i, turn in enumerate(history): user_msg, ai_msg = turn user_content = None if isinstance(user_msg, str): if not user_msg.startswith("[") and not user_msg.startswith("(改用文字輸入:") and not user_msg.startswith("(語音辨識結果:"): user_content = user_msg elif isinstance(user_msg, tuple): if i == len(history) - 1 and not user_text.startswith("["): user_content = user_text if user_content: llm_messages.append({"role": "user", "content": user_content}) ai_content = None if isinstance(ai_msg, str): if not ai_msg.startswith("["): ai_content = ai_msg elif isinstance(ai_msg, tuple) and len(ai_msg) > 0: if isinstance(ai_msg[0], str) and ai_msg[0].endswith(".mp3"): pass elif isinstance(ai_msg[0], str) and not ai_msg[0].startswith("["): ai_content = ai_msg[0] if ai_content: llm_messages.append({"role": "assistant", "content": ai_content}) if not llm_messages or llm_messages[-1]["role"] == "assistant": if not user_text.startswith("["): llm_messages.append({"role": "user", "content": user_text}) else: print("Skipping LLM call due to invalid final user text.") return history, None, "", final_topic ai_response_text = call_llm(final_topic, user_stance, llm_messages) ai_response_audio_path = call_tts(ai_response_text) last_user_turn_index = -1 for i in range(len(history) - 1, -1, -1): if history[i][1] is None and not history[i][0] is None: last_user_turn_index = i break if last_user_turn_index != -1: history[last_user_turn_index] = (history[last_user_turn_index][0], ai_response_text) if ai_response_audio_path and not ai_response_text.startswith("["): history.append((None, (ai_response_audio_path,))) else: print("Warning: Could not find user's turn. Appending AI response.") history.append(("[用戶回合丟失?]", ai_response_text)) if ai_response_audio_path and not ai_response_text.startswith("["): history.append((None, (ai_response_audio_path,))) return history, None, "", final_topic # --- Gradio UI (與上一版相同) --- with gr.Blocks(theme=gr.themes.Soft(), title="時事觀點對對碰 (OpenAI + 自訂主題)") as demo: gr.Markdown("## 🗣️ 時事觀點對對碰") gr.Markdown("選擇預設議題或輸入自訂議題，選擇立場，用文字或語音提出論點，AI 將扮演對手與你辯論！") chat_history = gr.State([]) with gr.Row(): topic_dd = gr.Dropdown(CURRENT_TW_TOPICS, label="選擇預設辯論主題", value=CURRENT_TW_TOPICS[0]) custom_topic_txt = gr.Textbox(label="或輸入自訂辯論主題", placeholder="若此處輸入，將優先使用此主題...") stance_radio = gr.Radio(["正方", "反方"], label="選擇你的立場", value="正方") chatbot_ui = gr.Chatbot(label="辯論區", height=500, render_markdown=True, bubble_full_width=False) with gr.Row(): with gr.Column(scale=7): user_txt = gr.Textbox(label="輸入你的論點 (文字)", placeholder="在此輸入文字...") with gr.Column(scale=3): user_audio = gr.Audio(sources=["microphone"], type="filepath", label="或錄製你的論點 (語音)") submit_btn = gr.Button("送出論點", variant="primary") # --- 事件綁定 --- submit_btn.click( fn=debate_turn, inputs=[topic_dd, custom_topic_txt, stance_radio, user_txt, user_audio, chatbot_ui], outputs=[chatbot_ui, user_audio, user_txt, custom_topic_txt] ) # 當選擇下拉選單時，清空自訂輸入框 def clear_custom_topic(dropdown_value): if dropdown_value: return "" return gr.skip() topic_dd.change(fn=clear_custom_topic, inputs=[topic_dd], outputs=[custom_topic_txt]) # 當在自訂輸入框打字時，清除下拉選單的選擇 def clear_dropdown(custom_text): if custom_text and custom_text.strip(): return None return gr.skip() # Corrected: Use lowercase 's' custom_topic_txt.change(fn=clear_dropdown, inputs=[custom_topic_txt], outputs=[topic_dd]) if __name__ == "__main__": demo.launch(debug=True)