feat: Create Effi-STT - custom STT solution for OpenWebUI

- FastAPI server with OpenAI-compatible API - Integration with faster-whisper (large-v3) - Full Docker support with docker-compose - Support for multiple languages (FR/EN/Auto) - Clean, maintainable codebase (~100 lines) - Health endpoint and error handling - Volume caching for models
2026-04-05 22:20:18 +02:00
parent eb2c365220
commit 7c0a0a39ba
5 changed files with 359 additions and 0 deletions
@@ -0,0 +1,138 @@
+# 🎙️ Effi-STT - Speech-to-Text pour Effi
+# API OpenAI-compatible avec faster-whisper
+
+from fastapi import FastAPI, File, UploadFile, Form, HTTPException
+from fastapi.responses import JSONResponse
+import uvicorn
+from faster_whisper import WhisperModel
+import tempfile
+import os
+from typing import Optional
+import logging
+
+# Configuration du logging
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+app = FastAPI(
+    title="Effi-STT",
+    description="Speech-to-Text API compatible OpenAI pour Effi",
+    version="1.0.0"
+)
+
+# Configuration
+MODEL_SIZE = os.getenv("MODEL_SIZE", "large-v3")
+DEVICE = os.getenv("DEVICE", "cpu")
+COMPUTE_TYPE = os.getenv("COMPUTE_TYPE", "int8")
+LANGUAGE = os.getenv("LANGUAGE", "fr")
+
+logger.info(f"🎙️ Chargement du modèle Whisper: {MODEL_SIZE}")
+logger.info(f"   Device: {DEVICE}, Compute type: {COMPUTE_TYPE}")
+
+# Chargement du modèle au démarrage
+model = WhisperModel(
+    MODEL_SIZE,
+    device=DEVICE,
+    compute_type=COMPUTE_TYPE,
+    cpu_threads=int(os.getenv("CPU_THREADS", "4"))
+)
+logger.info("✅ Modèle chargé avec succès!")
+
+@app.get("/health")
+async def health():
+    """Endpoint de vérification de santé"""
+    return {"status": "healthy", "model": MODEL_SIZE}
+
+@app.post("/v1/audio/transcriptions")
+async def transcribe(
+    file: UploadFile = File(..., description="Fichier audio à transcrire"),
+    model_param: Optional[str] = Form(None, alias="model"),
+    language: Optional[str] = Form(LANGUAGE, description="Code langue (fr, en, auto)"),
+    prompt: Optional[str] = Form(None, description="Prompt optionnel"),
+    response_format: Optional[str] = Form("json", description="Format de réponse")
+):
+    """
+    Endpoint de transcription compatible API OpenAI
+    """
+    try:
+        logger.info(f"📝 Transcription demandée: lang={language}, file={file.filename}")
+        
+        # Sauvegarder le fichier temporairement
+        suffix = os.path.splitext(file.filename)[1] or ".mp3"
+        with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
+            content = await file.read()
+            tmp.write(content)
+            tmp_path = tmp.name
+            logger.info(f"   Fichier sauvegardé: {tmp_path} ({len(content)} bytes)")
+        
+        try:
+            # Transcription avec faster-whisper
+            segments, info = model.transcribe(
+                tmp_path,
+                language=language if language != "auto" else None,
+                task="transcribe",
+                beam_size=5,
+                best_of=5,
+                condition_on_previous_text=True
+            )
+            
+            # Concaténer tous les segments
+            text_parts = []
+            for segment in segments:
+                text_parts.append(segment.text)
+            
+            full_text = " ".join(text_parts).strip()
+            
+            logger.info(f"✅ Transcription réussie: '{full_text[:50]}...'")
+            logger.info(f"   Langue détectée: {info.language}, Probabilité: {info.language_probability:.2f}")
+            
+            # Format OpenAI standard
+            if response_format == "text":
+                return full_text
+            elif response_format == "srt":
+                # TODO: Implémenter format SRT
+                return full_text
+            elif response_format == "vtt":
+                # TODO: Implémenter format VTT
+                return full_text
+            else:
+                # Format JSON (défaut)
+                return JSONResponse(content={"text": full_text})
+                
+        finally:
+            # Nettoyage
+            os.unlink(tmp_path)
+            
+    except Exception as e:
+        logger.error(f"❌ Erreur transcription: {str(e)}")
+        raise HTTPException(status_code=500, detail=str(e))
+
+@app.get("/v1/models")
+async def list_models():
+    """Liste les modèles disponibles (compatibilité OpenAI)"""
+    return {
+        "data": [
+            {
+                "id": "whisper-1",
+                "object": "model",
+                "created": 1677610602,
+                "owned_by": "openai",
+                "permission": [],
+                "root": "whisper-1",
+                "parent": None
+            }
+        ],
+        "object": "list"
+    }
+
+if __name__ == "__main__":
+    port = int(os.getenv("PORT", "8200"))
+    host = os.getenv("HOST", "0.0.0.0")
+    
+    logger.info(f"🚀 Démarrage du serveur sur http://{host}:{port}")
+    logger.info(f"   Endpoints:")
+    logger.info(f"   - GET  /health")
+    logger.info(f"   - POST /v1/audio/transcriptions")
+    logger.info(f"   - GET  /v1/models")
+    
+    uvicorn.run(app, host=host, port=port, log_level="info")