Update app.py

app.py

@@ -6,9 +6,10 @@ import csv
 import zipfile
 import re
 import difflib
+import tempfile
 from typing import List, Optional, Dict, Any
 
-from fastapi import FastAPI, UploadFile, File, HTTPException
+from fastapi import FastAPI, UploadFile, File, HTTPException, Form
 from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel
 
@@ -16,6 +17,7 @@ import torch
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 from langdetect import detect
 from transformers import MarianMTModel, MarianTokenizer
+from openai import OpenAI
 
 # ======================================================
 # 0) Configuración general
@@ -33,6 +35,12 @@ os.makedirs(UPLOAD_DIR, exist_ok=True)
 # { conn_id: { "db_path": str, "label": str } }
 DB_REGISTRY: Dict[str, Dict[str, Any]] = {}
 
+# Cliente OpenAI para transcripción de audio (Whisper / gpt-4o-transcribe)
+OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
+if not OPENAI_API_KEY:
+    print("⚠️ OPENAI_API_KEY no está definido. El endpoint /speech-infer no funcionará hasta configurarlo.")
+openai_client = OpenAI(api_key=OPENAI_API_KEY) if OPENAI_API_KEY else None
+
 # ======================================================
 # 1) Inicialización de FastAPI
 # ======================================================
@@ -641,6 +649,11 @@ class InferResponse(BaseModel):
     candidates: List[Dict[str, Any]]
 
 
+class SpeechInferResponse(BaseModel):
+    transcript: str
+    result: InferResponse
+
+
 # ======================================================
 # 7) Endpoints FastAPI
 # ======================================================
@@ -784,6 +797,57 @@ async def infer_sql(req: InferRequest):
     return InferResponse(**result)
 
 
+@app.post("/speech-infer", response_model=SpeechInferResponse)
+async def speech_infer(
+    connection_id: str = Form(...),
+    audio: UploadFile = File(...)
+):
+    """
+    Endpoint para consultas por VOZ:
+    - Recibe audio desde el navegador (multipart/form-data).
+    - Usa gpt-4o-transcribe para obtener el texto.
+    - Reutiliza el pipeline NL→SQL existente.
+    """
+    if openai_client is None:
+        raise HTTPException(
+            status_code=500,
+            detail="OPENAI_API_KEY no está configurado en el backend."
+        )
+
+    if audio.content_type is None:
+        raise HTTPException(status_code=400, detail="Archivo de audio inválido.")
+
+    # 1) Guardar audio temporalmente
+    try:
+        with tempfile.NamedTemporaryFile(delete=False, suffix=".webm") as tmp:
+            tmp.write(await audio.read())
+            tmp_path = tmp.name
+    except Exception:
+        raise HTTPException(status_code=500, detail="No se pudo procesar el audio recibido.")
+
+    # 2) Transcribir con gpt-4o-transcribe
+    try:
+        with open(tmp_path, "rb") as f:
+            transcription = openai_client.audio.transcriptions.create(
+                model="gpt-4o-transcribe",
+                file=f,
+                # language="es",  # opcional, si quieres forzar español
+            )
+        transcript_text: str = transcription.text
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Error al transcribir audio: {e}")
+
+    # 3) Reutilizar el pipeline NL→SQL con el texto transcrito
+    result_dict = nl2sql_with_rerank(transcript_text, connection_id)
+    infer_result = InferResponse(**result_dict)
+
+    # 4) Devolver transcripción + resultado NL→SQL
+    return SpeechInferResponse(
+        transcript=transcript_text,
+        result=infer_result,
+    )
+
+
 @app.get("/health")
 async def health():
     return {
@@ -804,6 +868,7 @@ async def root():
             "GET  /schema/{id}       (esquema resumido)",
             "GET  /preview/{id}/{t}  (preview de tabla)",
             "POST /infer             (NL→SQL + ejecución)",
+            "POST /speech-infer      (NL por voz → SQL + ejecución)",
             "GET  /health            (estado del backend)",
             "GET  /docs              (OpenAPI UI)",
         ],