feat: Migrate Speech Emotion Recognition to a dedicated Hugging Face Space and update client-side integration.

Dockerfile

@@ -0,0 +1,24 @@
+FROM python:3.9-slim
+
+WORKDIR /app
+
+# Instalar dependencias del sistema necesarias para audio y FunASR
+RUN apt-get update && apt-get install -y \
+    libsndfile1 \
+    ffmpeg \
+    && rm -rf /var/lib/apt/lists/*
+
+# Instalar dependencias Python
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copiar el código de la app
+COPY app /app
+
+# Removed build-time model caching to avoid OOM. Model will download on first boot.
+
+# Exponer el puerto de Hugging Face Spaces
+EXPOSE 7860
+
+# Comando para iniciar fastapi
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]

app/main.py

@@ -0,0 +1,91 @@
+from fastapi import FastAPI, UploadFile, File, HTTPException
+import shutil
+import os
+from funasr import AutoModel
+import uvicorn
+
+app = FastAPI(title="AInterviewer SER Model", version="emotion2vec_plus_large_interino")
+
+# Cargar modelo en memoria al inicio
+print("Cargando modelo emotion2vec_plus_large...")
+try:
+    model = AutoModel(model="iic/emotion2vec_plus_large")
+    print("Modelo cargado exitosamente.")
+except Exception as e:
+    print(f"Error cargando modelo: {e}")
+    model = None
+
+# Mapeo de 9 clases a 4 clases
+EMOTION_MAP = {
+    "happy": "happy",
+    "sad": "sad",
+    "angry": "angry",
+    "disgusted": "angry",
+    "neutral": "neutral",
+    "other": "neutral",
+    "unknown": "neutral",
+    "fearful": "neutral",  # o descartar
+    "surprised": "neutral" # o descartar
+}
+
+@app.get("/health")
+def healthcheck():
+    return {"status": "ok", "model_loaded": model is not None}
+
+@app.get("/version")
+def version():
+    return {"version": "emotion2vec_plus_large_interino"}
+
+@app.post("/analyze")
+async def analyze_audio(file: UploadFile = File(...)):
+    if not model:
+        raise HTTPException(status_code=503, detail="Model unvailable")
+
+    temp_path = f"/tmp/{file.filename}"
+    try:
+        with open(temp_path, "wb") as buffer:
+            shutil.copyfileobj(file.file, buffer)
+
+        # Inferencia
+        result = model.generate(temp_path, granularity="utterance", extract_embedding=False)
+        
+        # Result es una lista de dicts. Tomamos el primer elemento.
+        # Format ej: [{'labels': ['happy', 'neutral'], 'scores': [0.8, 0.2]}]
+        if not result or not isinstance(result, list):
+            raise ValueError("Unexpected model output format")
+            
+        res_data = result[0]
+        labels = res_data.get("labels", [])
+        scores = res_data.get("scores", [])
+        
+        # Sumar scores para nuestras 4 clases
+        mapped_scores = {"happy": 0.0, "sad": 0.0, "angry": 0.0, "neutral": 0.0}
+        
+        for label, score in zip(labels, scores):
+            target_class = EMOTION_MAP.get(label, "neutral")
+            mapped_scores[target_class] += float(score)
+            
+        # Normalizar si no suma 1 por el descarte
+        total = sum(mapped_scores.values())
+        if total > 0:
+            for k in mapped_scores:
+                mapped_scores[k] = round(mapped_scores[k] / total, 4)
+                
+        dominant = max(mapped_scores.items(), key=lambda x: x[1])[0]
+
+        return {
+            "dominant": dominant,
+            "distribution": mapped_scores,
+            "model": "emotion2vec_plus_large_interino",
+            "raw_labels": labels, # útil para debug
+            "raw_scores": [round(float(s), 4) for s in scores]
+        }
+
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+    finally:
+        if os.path.exists(temp_path):
+            os.remove(temp_path)
+
+if __name__ == "__main__":
+    uvicorn.run(app, host="0.0.0.0", port=7860)

requirements.txt

@@ -0,0 +1,6 @@
+fastapi
+uvicorn
+python-multipart
+funasr
+modelscope
+torchaudio