Update app.py

app.py

@@ -32,9 +32,9 @@ class ModelManager:
             cls._music_pipeline = pipeline("text-to-audio", "facebook/musicgen-small", device="cpu")
         return cls._music_pipeline
 
-# 멤버별 고유 성격/말투 페르소나
+# 페르소나 및 멤버 맵 (기존 유지)
 PERSONA_MAP = {
-    "서윤 (Korea)": "서정적이면서도 강렬한 K-Rock 리더. 진지하고 철학적이며 따뜻하게 조언함.",
+    "서윤 (Korea)": "서정적이면서도 강렬한 K-Rock 리더. 진지하고 따뜻하게 조언함.",
     "Chloe (USA)": "자유분방한 캘리포니아 펑크 락스타. 거침없고 쿨하며 에너제틱한 말투.",
     "Naomi (Japan)": "완벽주의 J-Rock 기타리스트. 섬세하고 예의 바르며 음악 이론에 날카로움.",
     "Beatrice (Brazil)": "열정적인 삼바 락커. 긍정적이고 리듬감이 넘치며 인생의 즐거움을 강조함.",
@@ -44,7 +44,6 @@ PERSONA_MAP = {
     "Sophie (France)": "예술적 자존심이 강한 아방가르드 락커. 시적이고 세련된 표현을 즐겨 씀."
 }
 
-# 언어별 멤버/보이스 매핑
 LANG_MEMBER_MAP = {
     "Korean": {"name": "서윤 (Korea)", "voice": "ko-KR-SunHiNeural"},
     "English": {"name": "Chloe (USA)", "voice": "en-US-AriaNeural"},
@@ -65,18 +64,18 @@ async def band_consulting(user_input, selected_lang, g_inst, b_inst, d_inst, cho
         m_info = LANG_MEMBER_MAP.get(selected_lang, LANG_MEMBER_MAP["Korean"])
         persona = PERSONA_MAP.get(m_info['name'], "열정적인 락스타")
         
-        # [정교화] 드럼 비트 + 타브 솔로 포함 프롬프트
+        # [지시 수정] 본문은 5줄 이내 요약, 나머지는 [TAB] 섹션에 통합
         system_prompt = f"""You are the rock star '{m_info['name']}'. Respond ONLY in {selected_lang}.
         Persona: {persona}.
 
         CRITICAL RULES:
-        1. Advice: 5-7 deep sentences in {selected_lang}. Speak according to your persona.
-        2. [TAB] SECTION: Must be highly detailed. 
-           - Include Drum patterns (e.g., H|x-x-x-x| S|--o---o-|).
-           - Include Guitar Solo Tab with fret numbers (e.g., e|--12-15-12--|).
-           - Provide precise Chord fingerings.
-        3. [TRANSLATION]: English translation of your advice.
-        4. [MUSIC]: English prompt: Guitar:{g_inst}, Bass:{b_inst}, Drums:{d_inst}, Chords:{chords}
+        1. MAIN ADVICE: Summarize your advice in exactly 3 to 5 sentences. This will be shown in the main chat.
+        2. [TAB] SECTION: This is where you provide all long, detailed explanations.
+           - Explain the theory, chord progression, and tips in detail here.
+           - Must include Drum patterns (e.g., H|x-x-x-x| S|--o---o-|).
+           - Must include Guitar Solo Tab with fret numbers (e.g., e|--12-15-12--|).
+        3. [TRANSLATION]: English translation of your 3-5 sentence advice only.
+        4. [MUSIC]: English prompt for MusicGen: Guitar:{g_inst}, Bass:{b_inst}, Drums:{d_inst}, Chords:{chords}
         """
         
         ai_text_raw = ""
@@ -96,15 +95,14 @@ async def band_consulting(user_input, selected_lang, g_inst, b_inst, d_inst, cho
             out = qwen(input_t, max_new_tokens=1024)
             ai_text_raw = out[0]['generated_text'].split("assistant\n")[-1]
 
-        # 파싱 로직
         tab_match = re.search(r'\[TAB\](.*?)(\[|$)', ai_text_raw, re.DOTALL | re.IGNORECASE)
         music_match = re.search(r'\[MUSIC:(.*?)\]', ai_text_raw, re.IGNORECASE)
         trans_match = re.search(r'\[TRANSLATION\](.*?)(\[|$)', ai_text_raw, re.DOTALL | re.IGNORECASE)
         
-        tab_display = tab_match.group(1).strip() if tab_match else "No Detailed Tab Generated"
+        tab_display = tab_match.group(1).strip() if tab_match else "No Detailed Data"
         eng_translation = trans_match.group(1).strip() if trans_match else ""
 
-        # [수정] TTS 음성에서 번역문 완벽 제외 로직
+        # TTS 음성에서 번역문 제외 로직 강화
         speech_text = re.sub(r'\[TAB\].*?(\[|$)', '', ai_text_raw, flags=re.DOTALL | re.IGNORECASE)
         speech_text = re.sub(r'\[MUSIC:.*?\]', '', speech_text, flags=re.IGNORECASE)
         speech_text = re.sub(r'\[TRANSLATION\].*?(\[|$)', '', speech_text, flags=re.DOTALL | re.IGNORECASE).strip()
@@ -113,9 +111,8 @@ async def band_consulting(user_input, selected_lang, g_inst, b_inst, d_inst, cho
         communicate = edge_tts.Communicate(tts_text, m_info["voice"])
         await communicate.save(voice_path)
 
-        # MusicGen 연주 생성 (12초)
         music_gen = ModelManager.get_music()
-        music_p = music_match.group(1).strip() if music_match else "rock music"
+        music_p = music_match.group(1).strip() if music_match else "rock"
         music_output = music_gen(music_p, forward_params={"max_new_tokens": 512})
         audio_data = np.squeeze(music_output["audio"])
         audio_int16 = (audio_data * 32767).astype(np.int16)
@@ -124,7 +121,7 @@ async def band_consulting(user_input, selected_lang, g_inst, b_inst, d_inst, cho
         return speech_text, voice_path, music_path, tab_display, eng_translation
 
     except Exception as e:
-        return f"Error: {str(e)}", None, None, "No Tab", "Error occurred"
+        return f"Error: {str(e)}", None, None, "No Data", "Error"
 
 with gr.Blocks() as demo:
     inputs = [gr.Textbox(visible=False) for _ in range(6)]