Update app.py

app.py

@@ -8,102 +8,64 @@ import scipy.io.wavfile
 import numpy as np
 from transformers import pipeline
 
-# 1. API 키 설정 (Hugging Face Settings -> Variables에 등록 필수)
+# 1. API 키 설정
 GENAI_KEY = os.getenv("GEMINI_KEY")
 
-# 2. [작곡가] MusicGen 모델 로드 (음질 보정 설정)
-print("⏳ 작곡가(MusicGen)를 소집 중입니다... (1~2분 소요)")
+# 2. [작곡가] MusicGen 로드
+print("⏳ 작곡가 소집 중...")
 try:
-    # facebook/musicgen-small 모델 사용
     music_synthesiser = pipeline("text-to-audio", "facebook/musicgen-small")
-    print("✅ 작곡가 준비 완료! 잼 세션을 시작할 수 있습니다.")
+    print("✅ 작곡가 준비 완료!")
 except Exception as e:
     print(f"⚠️ 모델 로드 실패: {e}")
     music_synthesiser = None
 
-# 3. 최적의 Gemini 모델 자동 탐색
 def find_working_model():
     url = f"https://generativelanguage.googleapis.com/v1beta/models?key={GENAI_KEY}"
     try:
         response = requests.get(url)
         data = response.json()
         candidates = [m['name'].replace("models/", "") for m in data.get('models', []) if 'generateContent' in m.get('supportedGenerationMethods', [])]
-        preferred = ["gemini-1.5-flash", "gemini-1.5-flash-latest", "gemini-pro"]
+        preferred = ["gemini-1.5-flash", "gemini-pro"]
         for p in preferred:
             if p in candidates: return p
         return candidates[0] if candidates else "gemini-pro"
-    except:
-        return "gemini-pro"
+    except: return "gemini-pro"
 
 ACTIVE_MODEL = find_working_model()
 
-# 4. 밴드 멤버 데이터 (이름 오타 완벽 수정: "서윤 (Korea)")
+# 8인 멤버 데이터
 MEMBERS = {
-    "서윤 (Korea)": { 
-        "voice": "ko-KR-SunHiNeural", 
-        "prompt": "당신은 록밴드 보컬 '서윤'입니다. 한국어로 대답하세요. 성격: 시니컬, 츤데레, 음악에 진심. 전문적인 음악 조언을 제공하세요." 
-    },
-    "Chloe (USA)": { 
-        "voice": "en-US-AriaNeural", 
-        "prompt": "You are 'Chloe', the lead guitarist. Respond in English. Focus on guitar tone and riffs." 
-    },
-    "Beatrice (Brazil)": { 
-        "voice": "pt-BR-FranciscaNeural", 
-        "prompt": "You are 'Beatrice', the drummer. Focus on rhythm patterns and energy." 
-    },
-    "Naomi (Japan)": { 
-        "voice": "ja-JP-NanamiNeural", 
-        "prompt": "You are 'Naomi'. Respond in Japanese. Focus on harmony and chord progressions." 
-    },
-    "Elena (Spain)": { 
-        "voice": "es-ES-ElviraNeural", 
-        "prompt": "You are 'Elena'. Focus on bass lines and groove." 
-    },
-    "Amira (Egypt)": { 
-        "voice": "ar-EG-SalmaNeural", 
-        "prompt": "You are 'Amira'. Focus on keyboards and atmosphere." 
-    },
-    "Liwei (China)": { 
-        "voice": "zh-CN-XiaoxiaoNeural", 
-        "prompt": "You are 'Liwei'. Focus on electronic sound and FX." 
-    },
-    "Sophie (France)": { 
-        "voice": "fr-FR-DeniseNeural", 
-        "prompt": "You are 'Sophie'. Focus on emotional melodies and violin." 
-    }
+    "서윤 (Korea)": { "voice": "ko-KR-SunHiNeural", "prompt": "당신은 보컬 '서윤'입니다. 시니컬하고 음악에 진심입니다." },
+    "Chloe (USA)": { "voice": "en-US-AriaNeural", "prompt": "You are 'Chloe', the lead guitarist. Cool rockstar vibe." },
+    "Beatrice (Brazil)": { "voice": "pt-BR-FranciscaNeural", "prompt": "You are 'Beatrice', the drummer. Focus on rhythm." },
+    "Naomi (Japan)": { "voice": "ja-JP-NanamiNeural", "prompt": "You are 'Naomi'. Focus on harmony." },
+    "Elena (Spain)": { "voice": "es-ES-ElviraNeural", "prompt": "You are 'Elena'. Focus on groove." },
+    "Amira (Egypt)": { "voice": "ar-EG-SalmaNeural", "prompt": "You are 'Amira'. Focus on atmosphere." },
+    "Liwei (China)": { "voice": "zh-CN-XiaoxiaoNeural", "prompt": "You are 'Liwei'. Focus on sound design." },
+    "Sophie (France)": { "voice": "fr-FR-DeniseNeural", "prompt": "You are 'Sophie'. Focus on melody." }
 }
 
-# 5. 메인 로직 함수
 async def band_consulting(user_input, member_name, consult_category):
     try:
         member = MEMBERS.get(member_name, MEMBERS["서윤 (Korea)"])
         
-        # ★ 언어 결정 로직 보강: 카테고리 텍스트에서 언어 정보를 확실히 추출
-        # 예: "🇺🇸 English (Global Random 🎲) / Guitar" -> "English"
+        # 언어 강제 로직
         requested_lang = "English" if "English" in consult_category else "Korean"
 
         system_instruction = f"""
         {member['prompt']}
+        [CRITICAL RULE] Respond ENTIRELY in {requested_lang}. 
+        Even if you are global, the user selected {requested_lang} as the consultation language.
         
-        [CRITICAL RULE]
-        The user wants to consult in {requested_lang}. 
-        Regardless of your origin, you MUST respond ENTIRELY in {requested_lang}.
-        Do not use your native language if it's different from {requested_lang}.
-        
-        [Current Category: {consult_category}]
-        
-        조언을 제공한 뒤 반드시 마��막에 다음 형식을 포함하세요:
-        1. [TAB]: 음악의 간단한 코드 진행이나 텍스트 악보를 그리세요.
-        2. [MUSIC]: 분야와 어울리는 고음질 음악 프롬프트를 영어로 작성하세요.
+        조언 후 반드시 포함:
+        1. [TAB]: 코드 진행 또는 텍스트 악보.
+        2. [MUSIC]: 분야와 어울리는 영어 음악 프롬프트.
         """
         
         url = f"https://generativelanguage.googleapis.com/v1beta/models/{ACTIVE_MODEL}:generateContent?key={GENAI_KEY}"
         payload = {"contents": [{"parts": [{"text": f"{system_instruction}\n질문: {user_input}"}]}]}
         res = requests.post(url, json=payload, headers={'Content-Type': 'application/json'})
-        
-        if res.status_code != 200:
-            return "Gemini API 연결에 실패했습니다.", None, None, "No Tab"
-            
         ai_text_raw = res.json()['candidates'][0]['content']['parts'][0]['text']
         
         tab_match = re.search(r'\[TAB\](.*?)(\[|$)', ai_text_raw, re.DOTALL)
@@ -112,42 +74,26 @@ async def band_consulting(user_input, member_name, consult_category):
         
         clean_text = re.sub(r'\[TAB\].*?(\[|$)', '', ai_text_raw, flags=re.DOTALL)
         clean_text = re.sub(r'\[MUSIC:.*?\]', '', clean_text).strip()
-        tts_text = re.sub(r'[\*\#\-\_\~]', '', clean_text) 
+        tts_text = re.sub(r'[\*\#\-\_\~]', '', clean_text) # 특수문자 제거
 
         voice_path = f"/tmp/v_{member_name.replace(' ', '_')}.mp3"
         await edge_tts.Communicate(tts_text, member['voice']).save(voice_path)
         
         music_path = None
         if music_match and music_synthesiser:
-            music_p = music_match.group(1).strip() + ", high quality, studio recording, power rock, intense drums"
+            music_p = music_match.group(1).strip() + ", high quality, studio recording, master sound"
             music_output = music_synthesiser(music_p, forward_params={"max_new_tokens": 512, "guidance_scale": 4.5, "do_sample": True})
             music_path = f"/tmp/m_{member_name.replace(' ', '_')}.wav"
             scipy.io.wavfile.write(music_path, music_output["sampling_rate"], music_output["audio"][0].T)
 
         return clean_text, voice_path, music_path, tab_display
-
     except Exception as e:
-        print(f"❌ Error: {e}")
-        return f"밴드 시스템 오류: {str(e)}", None, None, "Error"
+        return f"Error: {str(e)}", None, None, "Error"
 
-# 8. Gradio 인터페이스 정의 (API용)
 with gr.Blocks() as demo:
     with gr.Row():
-        inp_text = gr.Textbox()
-        inp_member = gr.Textbox()
-        inp_cat = gr.Textbox()
-    
-    out_advice = gr.Textbox()
-    out_voice = gr.Audio()
-    out_music = gr.Audio()
-    out_tab = gr.Textbox()
-    
-    btn = gr.Button("Predict")
-    btn.click(
-        fn=band_consulting, 
-        inputs=[inp_text, inp_member, inp_cat], 
-        outputs=[out_advice, out_voice, out_music, out_tab],
-        api_name="predict"
-    )
+        i1 = gr.Textbox(); i2 = gr.Textbox(); i3 = gr.Textbox()
+    o1 = gr.Textbox(); o2 = gr.Audio(); o3 = gr.Audio(); o4 = gr.Textbox()
+    btn = gr.Button("GO"); btn.click(band_consulting, [i1, i2, i3], [o1, o2, o3, o4], api_name="predict")
 
 demo.launch()