Update app.py

app.py

@@ -7,6 +7,7 @@ import json
 import time
 import subprocess
 import logging
+import numpy as np
 from flask import Flask, request, jsonify, render_template_string, send_from_directory
 import whisper
 import edge_tts
@@ -23,10 +24,18 @@ os.makedirs(UPLOAD_FOLDER, exist_ok=True)
 
 tasks = {}
 
+# --- MAP SUARA (MALE & FEMALE) ---
 VOICE_MAP = {
-    'id-ID': 'id-ID-ArdiNeural', 
-    'en-US': 'en-US-ChristopherNeural', 
-    'ja-JP': 'ja-JP-KeitaNeural'
+    'id-ID': {'Male': 'id-ID-ArdiNeural', 'Female': 'id-ID-GadisNeural'}, 
+    'en-US': {'Male': 'en-US-ChristopherNeural', 'Female': 'en-US-AriaNeural'}, 
+    'ja-JP': {'Male': 'ja-JP-KeitaNeural', 'Female': 'ja-JP-NanamiNeural'}
+}
+
+# Mapping Bahasa untuk Prompt AI
+LANG_MAP = {
+    'id-ID': 'Indonesia',
+    'en-US': 'Inggris',
+    'ja-JP': 'Jepang'
 }
 
 # Load Whisper (CPU Friendly, FP16 Fixed)
@@ -38,10 +47,54 @@ def get_audio_duration(file_path):
         '-of', 'default=noprint_wrappers=1:nokey=1', file_path
     ]
     result = subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.DEVNULL)
-    return float(result.stdout)
+    try:
+        return float(result.stdout)
+    except:
+        return 0.0
 
-def translate_segments_llm(segments, custom_prompt):
-    instruction = custom_prompt if custom_prompt else "Terjemahkan teks dalam JSON ini. Balas HANYA JSON array."
+def analyze_gender_and_pitch(audio_path):
+    """Menganalisis potongan audio untuk menentukan gender dan variasi pitch."""
+    try:
+        import librosa
+        # Load audio dengan sample rate standard
+        y, sr = librosa.load(audio_path, sr=22050)
+        
+        if len(y) == 0: return "Male", "+0Hz"
+
+        # Deteksi Fundamental Frequency (F0)
+        f0 = librosa.yin(y, fmin=65, fmax=300)
+        valid_f0 = f0[~np.isnan(f0)]
+        
+        if len(valid_f0) > 0:
+            mean_f0 = np.mean(valid_f0)
+            
+            # Threshold umum: > 165Hz = Perempuan, < 165Hz = Laki-laki
+            gender = "Female" if mean_f0 >= 165 else "Male"
+            
+            # Hitung variasi pitch (agar tiap orang suaranya beda)
+            # Normal cowok ~120Hz, cewek ~210Hz. Dibagi 2 agar tidak terlalu ekstrem
+            base_f0 = 210.0 if gender == "Female" else 120.0
+            pitch_shift = int((mean_f0 - base_f0) / 2) 
+            
+            # Batasi modifikasi pitch Edge TTS agar tidak rusak (antara -20Hz s/d +20Hz)
+            pitch_shift = max(-20, min(20, pitch_shift))
+            pitch_str = f"+{pitch_shift}Hz" if pitch_shift >= 0 else f"{pitch_shift}Hz"
+            
+            return gender, pitch_str
+    except Exception as e:
+        print(f"Pitch analysis warning: {e}")
+        
+    return "Male", "+0Hz" # Default fallback
+
+def translate_segments_llm(segments, custom_prompt, target_voice):
+    target_lang = LANG_MAP.get(target_voice, 'Indonesia')
+    
+    # PERBAIKAN: Memasukkan bahasa target secara paksa ke dalam prompt
+    if custom_prompt:
+        instruction = f"{custom_prompt}\n\nPENTING: Terjemahkan SEMUA teks ke dalam bahasa {target_lang}."
+    else:
+        instruction = f"Terjemahkan teks dalam JSON ini ke bahasa {target_lang} dengan akurat. Balas HANYA dengan JSON array."
+        
     input_data = [{"id": i, "text": s['text']} for i, s in enumerate(segments)]
     full_prompt = f"{instruction}\n\nFormat: [{{'id': 0, 'text': '...'}}]\n\nData:\n{json.dumps(input_data)}"
 
@@ -63,12 +116,14 @@ def translate_segments_llm(segments, custom_prompt):
         translated_list = json.loads(full_text[start_idx:end_idx])
         for item in translated_list:
             segments[item['id']]['translated_text'] = item['text']
-    except:
+    except Exception as e:
+        print(f"Translation Error: {e}")
         for s in segments: s['translated_text'] = s['text']
     return segments
 
-async def generate_tts(text, voice, path):
-    communicate = edge_tts.Communicate(text, voice)
+# PERBAIKAN: Menambahkan parameter pitch
+async def generate_tts(text, voice, path, pitch_str="+0Hz"):
+    communicate = edge_tts.Communicate(text, voice, pitch=pitch_str)
     await communicate.save(path)
 
 def process_dubbing(task_id, video_path, target_voice, custom_prompt):
@@ -81,10 +136,13 @@ def process_dubbing(task_id, video_path, target_voice, custom_prompt):
         result = whisper_model.transcribe(orig_audio, verbose=False, fp16=False)
         segments = result['segments']
 
-        tasks[task_id]['status'] = 'Translasi AI...'
-        translated_segments = translate_segments_llm(segments, custom_prompt)
+        tasks[task_id]['status'] = f'Translasi AI ({LANG_MAP.get(target_voice, target_voice)})...'
+        # Pass target_voice ke translator
+        translated_segments = translate_segments_llm(segments, custom_prompt, target_voice)
 
+        tasks[task_id]['status'] = 'Menganalisis Suara & Dubbing...'
         processed_audio_files = []
+        
         for i, seg in enumerate(translated_segments):
             start_t = seg['start']
             end_t = seg['end']
@@ -92,10 +150,21 @@ def process_dubbing(task_id, video_path, target_voice, custom_prompt):
             text = seg.get('translated_text', seg['text'])
             if not text.strip(): continue
 
+            # Potong audio asli khusus untuk segmen ini guna deteksi suara
+            chunk_wav = os.path.join(app.config['UPLOAD_FOLDER'], f"{task_id}_chunk_{i}.wav")
+            subprocess.run(['ffmpeg', '-loglevel', 'quiet', '-y', '-i', orig_audio, '-ss', str(start_t), '-t', str(duration_orig), chunk_wav], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+
+            # Deteksi Cewek/Cowok dan variasi pitch
+            gender, pitch_str = analyze_gender_and_pitch(chunk_wav)
+            
+            # Pilih Voice ID yang sesuai berdasarkan bahasa dan gender
+            selected_voice = VOICE_MAP.get(target_voice, VOICE_MAP['id-ID'])[gender]
+
             raw_tts = os.path.join(app.config['UPLOAD_FOLDER'], f"{task_id}_raw_{i}.mp3")
             sync_tts = os.path.join(app.config['UPLOAD_FOLDER'], f"{task_id}_sync_{i}.wav")
             
-            asyncio.run(generate_tts(text, VOICE_MAP.get(target_voice, 'id-ID-ArdiNeural'), raw_tts))
+            # Generate TTS dengan pitch modifier
+            asyncio.run(generate_tts(text, selected_voice, raw_tts, pitch_str))
             
             tts_dur = get_audio_duration(raw_tts)
             speed = min(max(tts_dur / duration_orig, 0.7), 1.8) if duration_orig > 0 else 1.0
@@ -108,8 +177,6 @@ def process_dubbing(task_id, video_path, target_voice, custom_prompt):
         output_path = os.path.join(app.config['UPLOAD_FOLDER'], output_filename)
         
         # LOGIKA AUDIO BARU:
-        # 1. Background (Video asli): Turunkan frekuensi vokal (-15dB di 1000Hz) & Set volume ke 40% (0.4) agar backsound tetap ada.
-        # 2. TTS Dubbing AI: Besarkan volumenya ke 300% (3.0) agar sangat jelas.
         filter_complex = "[0:a]equalizer=f=1000:width_type=o:w=2:g=-15,volume=0.4[bg];"
         inputs_cmd = ['ffmpeg', '-loglevel', 'quiet', '-y', '-i', video_path]
         amix_inputs = "[bg]"
@@ -118,11 +185,9 @@ def process_dubbing(task_id, video_path, target_voice, custom_prompt):
             idx = i + 1
             inputs_cmd.extend(['-i', item['path']])
             start_ms = int(item['start'] * 1000)
-            # Beri delay, dan besarkan volume TTS 3x lipat (300%)
             filter_complex += f"[{idx}:a]adelay={start_ms}|{start_ms},volume=3.0[dub{idx}];"
             amix_inputs += f"[dub{idx}]"
         
-        # Gabungkan semua, tambah volume akhir sedikit untuk kompensasi penurunan dari filter amix
         filter_complex += f"{amix_inputs}amix=inputs={len(processed_audio_files)+1}:duration=first:dropout_transition=0,volume=1.5[outa]"
         
         final_cmd = inputs_cmd + [
@@ -145,6 +210,7 @@ def process_dubbing(task_id, video_path, target_voice, custom_prompt):
         tasks[task_id]['status'] = 'Error'
         tasks[task_id]['error_message'] = str(e)
 
+
 # --- ROUTES ---
 
 @app.route('/')
@@ -184,7 +250,7 @@ HTML_TEMPLATE = """
 
     <div class="bg-gray-800 rounded-2xl shadow-2xl p-8 w-full max-w-md border border-gray-700">
         <h2 class="text-2xl font-bold text-center mb-2 text-white">🎙️ Dubbing Sync Pro</h2>
-        <p class="text-sm text-center text-gray-400 mb-6">Suara AI 300% | Backsound Asli 40%</p>
+        <p class="text-sm text-center text-gray-400 mb-6">Deteksi Gender & Multi-Speaker Auto-Pitch</p>
 
         <form id="uploadForm" class="space-y-4">
             <div>