feat: Complete optimization with 3 bugs fixed + backend-only

Dockerfile

@@ -25,12 +25,12 @@ RUN mkdir -p /app/uploads /app/models /app/frontend/dist
 RUN chmod -R 777 /app
 
 # Upgrade pip and install build tools
-RUN pip install --no-cache-dir --upgrade pip setuptools wheel
+RUN pip install --no-cache-dir --upgrade pip setuptools wheel Cython
 
-# Install Torch CPU specifically (Prevents Erro 500 and OOM)
+# Install Torch CPU specifically (Essential for HF Spaces CPU tier)
 RUN pip install --no-cache-dir torch==2.5.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cpu
 
-# Install backend dependencies
+# Install remaining dependencies (WhisperX will pull its required pyannote version here)
 COPY backend/requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 

backend/main.py

@@ -40,7 +40,11 @@ from fastapi.staticfiles import StaticFiles
 from fastapi.responses import JSONResponse, FileResponse
 from processor import TranscriptionProcessor
 
+import asyncio
+
 app = FastAPI()
+# Lock para evitar que múltiplas requisições fritem a CPU/RAM simultaneamente
+process_lock = asyncio.Lock()
 
 # Configuração de CORS Universal
 app.add_middleware(
@@ -61,69 +65,59 @@ async def process_media(
     script: UploadFile = File(None)
 ):
     session_id = uuid.uuid4().hex[:8]
-    print(f"\n🚀 [SESSÃO {session_id}] - INICIANDO PROCESSO")
-    
     audio_path = os.path.join(UPLOAD_DIR, f"{session_id}_{audio.filename}")
     script_path = None
     
+    # Garantir limpeza mesmo em erro fatal
     try:
-        # Salvamento de Áudio Robusto
+        # Salvamento
         with open(audio_path, "wb") as buffer:
             shutil.copyfileobj(audio.file, buffer)
         
-        script_text = ""
-        if script:
-            script_path = os.path.join(UPLOAD_DIR, f"{session_id}_{script.filename}")
-            with open(script_path, "wb") as buffer:
-                shutil.copyfileobj(script.file, buffer)
-            script_text = processor.process_docx(script_path)
-        
-        # TRANSCRICÃO COM SEGURANÇA TOTAL
-        words = processor.transcribe(audio_path, language="pt")
-        
-        # CORREÇÃO INTELIGENTE (Script Based)
-        if script_text:
-            words = processor.align_with_script(words, script_text)
+        async with process_lock: # FILA DE ESPERA INTELIGENTE
+            print(f"🚀 [SESSÃO {session_id}] Iniciando processamento exclusivo...")
             
-        words = processor.correct_orthography(words)
-        
-        # GERAÇÃO DO JSON
-        result_json = processor.generate_json(words)
-        json_filename = f"{session_id}_transcription.json"
-        json_path = os.path.join(UPLOAD_DIR, json_filename)
-        
-        # Escrita com persistência
-        with open(json_path, "w", encoding="utf-8") as f:
-            json.dump(result_json, f, ensure_ascii=False, indent=2)
-
-        # Cleanup de arquivos pesados (MANTÉM APENAS O JSON)
-        background_tasks.add_task(os.remove, audio_path)
-        if script_path and os.path.exists(script_path):
-            background_tasks.add_task(os.remove, script_path)
-        
-        # Limpeza de Memória RAM agendada
-        gc.collect()
+            script_text = ""
+            if script:
+                script_path = os.path.join(UPLOAD_DIR, f"{session_id}_{script.filename}")
+                with open(script_path, "wb") as buffer:
+                    shutil.copyfileobj(script.file, buffer)
+                script_text = processor.process_docx(script_path)
+            
+            # 1. Transcrição
+            words = processor.transcribe(audio_path, language="pt")
+            
+            # 2. Correção de Roteiro (Fuzzy)
+            if script_text:
+                words = processor.align_with_script(words, script_text)
+            
+            # 3. Limpeza Final
+            words = processor.correct_orthography(words)
+            result_json = processor.generate_json(words)
+            
+            json_filename = f"{session_id}_transcription.json"
+            json_path = os.path.join(UPLOAD_DIR, json_filename)
+            with open(json_path, "w", encoding="utf-8") as f:
+                json.dump(result_json, f, ensure_ascii=False, indent=2)
 
-        print(f"✅ [SESSÃO {session_id}] - FINALIZADA COM SUCESSO\n")
-        return {
-            "success": True,
-            "session_id": session_id,
-            "words": words,
-            "script_text": script_text,
-            "json_url": f"/api/download/{json_filename}"
-        }
+            return {
+                "success": True,
+                "session_id": session_id,
+                "words": words,
+                "json_url": f"/api/download/{json_filename}"
+            }
 
     except Exception as e:
-        err_msg = traceback.format_exc()
-        print(f"❌ [ERRO NA SESSÃO {session_id}]:\n{err_msg}")
-        return JSONResponse(
-            status_code=500, 
-            content={
-                "success": False, 
-                "error": str(e), 
-                "traceback": err_msg
-            }
-        )
+        print(f"❌ [ERRO SESSÃO {session_id}]: {str(e)}")
+        return JSONResponse(status_code=500, content={"success": False, "error": str(e)})
+    
+    finally:
+        # LIMPEZA OBRIGATÓRIA DE ÁUDIO (Anti-Leak)
+        if os.path.exists(audio_path):
+            os.remove(audio_path)
+        if script_path and os.path.exists(script_path):
+            os.remove(script_path)
+        gc.collect() # Garante liberação de RAM
 
 @app.get("/api/download/{filename}")
 async def download_json(filename: str):

backend/requirements.txt

@@ -7,5 +7,6 @@ pandas==2.2.2
 onnxruntime
 accelerate
 nest_asyncio
-pyannote.audio==3.3.1
-git+https://github.com/m-bain/whisperX.git@v3.1.1
+pydub
+librosa
+git+https://github.com/m-bain/whisperX.git

google_colab/colab_app.py

@@ -5,35 +5,38 @@ import os
 import sys
 import subprocess
 import difflib
+import time # Added for time.sleep
 
 def install_safe_stack():
     print("🛠️ LIMPANDO E CURANDO AMBIENTE (Aguarde 3 min)...")
-    
     try:
-        # 1. Limpeza Radical para evitar conflitos de versões "sequestradas"
+        # 1. Limpeza pesada
         print("🧹 Removendo versões instáveis...")
-        subprocess.check_call([sys.executable, "-m", "pip", "uninstall", "-y", "torch", "torchaudio", "torchvision", "whisperx", "pandas"])
+        subprocess.check_call([sys.executable, "-m", "pip", "uninstall", "-y", "torch", "torchvision", "torchaudio", "whisperx", "pandas"])
         
-        # 2. Instalação Sincronizada (A "Santíssima Trindade" estável para T4)
-        print("📦 Instalando PyTorch Stack Estável (2.5.1)...")
+        # 2. Instalação do Stack de Áudio/Vídeo (Caminho oficial CUDA 12.1)
+        print("📦 Instalando Motores CUDA...")
         subprocess.check_call([
-            sys.executable, "-m", "pip", "install",
+            sys.executable, "-m", "pip", "install", 
             "torch==2.5.1+cu121", "torchvision==0.20.1+cu121", "torchaudio==2.5.1+cu121",
-            "pandas==2.2.2", # Versão que o Colab exige
-            "--index-url", "https://download.pytorch.org/whl/cu121"
+            "--extra-index-url", "https://download.pytorch.org/whl/cu121"
         ])
         
-        # 3. WhisperX v3.1.1 (A versão mais estável já feita)
-        print("📦 Instalando WhisperX v3.1.1...")
-        subprocess.check_call([sys.executable, "-m", "pip", "install", "git+https://github.com/m-bain/whisperX.git@v3.1.1"])
-        
-        # 4. Dependências cruciais
-        print("📦 Finalizando componentes...")
-        subprocess.check_call([sys.executable, "-m", "pip", "install", "pyannote.audio==3.3.1", "gradio", "python-docx", "transformers", "accelerate", "nest_asyncio"])
+        # 3. Instalação de dependências de sistema e Python
+        print("📦 Instalando WhisperX e Pandas...")
+        subprocess.check_call([
+            sys.executable, "-m", "pip", "install",
+            "pandas==2.2.2", "pyannote.audio==3.3.1",
+            "git+https://github.com/m-bain/whisperX.git@v3.1.1",
+            "gradio", "python-docx", "transformers", "accelerate", "nest_asyncio" # Kept from original
+        ])
+
+        # 4. Dependências de sistema
+        print("📦 Finalizando componentes de sistema...")
         subprocess.check_call(["apt-get", "install", "-y", "-qq", "ffmpeg", "libsndfile1"])
         
-        print("\n✅ AMBIENTE CURADO COM SUCESSO!")
-        print("⚠️ AÇÃO NECESSÁRIA: Vá em 'Ambiente de Execução' > 'Reiniciar sessão' e rode esta célula de novo.")
+        print("\n✅ AMBIENTE CURADO! REINICIANDO PARA APLICAR...")
+        time.sleep(2)
         os.kill(os.getpid(), 9)
     except Exception as e:
         print(f"❌ Erro na cura: {e}")