src/expon/presentation/interfaces/rest/controllers/multimodal_controller.py

# src/expon/presentation/interfaces/rest/controllers/multimodal_controller.py
from fastapi import APIRouter, UploadFile, File, Form, HTTPException, Depends
from pydantic import BaseModel
from typing import Optional, Dict
import os
from tempfile import NamedTemporaryFile

from src.expon.presentation.domain.services.multimodal_service import MultimodalService

# 🔐 Usa la MISMA dependencia de auth del backend
from src.expon.iam.infrastructure.authorization.sfs.auth_bearer import get_current_user

router = APIRouter()
_service = MultimodalService()

class PredictResponse(BaseModel):
    dominant_emotion: str
    emotion_probabilities: Dict[str, float]
    transcript: Optional[str]  # ← puede no haber texto
    alpha: float
    temp_text: float
    provider: str

@router.post("/predict", response_model=PredictResponse)
async def predict_multimodal(
    audio: UploadFile = File(...),
    text: Optional[str] = Form(
        None,
        description="Texto opcional para fusión. Déjalo vacío para usar solo audio.",
        example=None,
    ),
    user=Depends(get_current_user),  # ← exige login con tu flujo real
):
    # 🔹 Sanitiza el placeholder de Swagger y vacíos
    if text is not None:
        t = text.strip()
        if not t or t.lower() == "string":
            text = None

    ext = (os.path.splitext(audio.filename or "")[-1] or "").lower()
    if ext not in [".wav", ".mp3", ".m4a"]:
        raise HTTPException(status_code=400, detail="Formato no soportado (.wav, .mp3, .m4a)")

    with NamedTemporaryFile(delete=False, suffix=ext) as tmp:
        tmp.write(await audio.read())
        tmp_path = tmp.name

    try:
        result = _service.predict(tmp_path, text)
        return PredictResponse(**result)
    finally:
        try:
            os.remove(tmp_path)
        except Exception:
            pass