voice module

spendoo/categorization/routes.py

@@ -1,8 +1,8 @@
+from spendoo.core.services import ServiceContainer
 from .models import CategorizationRequest, TransactionExtractionResponse
-from spendoo.categorization.service import CategorizationService
 from fastapi import APIRouter
 
-service = CategorizationService()
+service = ServiceContainer.get_categorization_service()
 router = APIRouter(prefix="/categorization", tags=["categorization"])
 
 

spendoo/categorization/service.py

@@ -26,8 +26,9 @@ def extract(self, text: str):
 
         For each item try to return:
         - id (incremental starting from 1 in order of appearance)
-        - item_name
+        - item_name exactly as it appears in the text (avoid interpreting it, just extract the name) but without any additional descriptions or measurements if they are mentioned in the same line.
         - price 
+        - category (choose one category from the list below that best fits the item based on its description, if you are unsure make it null)
         - category_id (choose one id from the list below)
 
         Available categories:

spendoo/core/config.py

@@ -6,4 +6,7 @@
 class Settings:
     GROQ_API_KEY: str = os.getenv("GROQ_API_KEY")   
     MISTRAL_API_KEY: str = os.getenv("MISTRAL_API_KEY")
+    VOICE_ALLOWED_EXTENSIONS: set[str] = {".wav", ".mp3", ".ogg", ".m4a", ".flac", ".webm"}
+    VOICE_MAX_SIZE: int = 25 * 1024 * 1024
+
 settings = Settings()

spendoo/core/services.py

@@ -0,0 +1,26 @@
+from spendoo.ocr.service import OCRService
+from spendoo.categorization.service import CategorizationService
+from spendoo.voice.service import VoiceService
+
+
+class ServiceContainer:
+    _voice_service = None
+    _ocr_service = None
+    _categorization_service = None
+    @classmethod
+    def get_voice_service(cls) -> VoiceService:
+        if cls._voice_service is None:
+            cls._voice_service = VoiceService()
+        return cls._voice_service
+
+    @classmethod
+    def get_ocr_service(cls) -> OCRService:
+        if cls._ocr_service is None:
+            cls._ocr_service = OCRService()
+        return cls._ocr_service
+
+    @classmethod
+    def get_categorization_service(cls) -> CategorizationService:
+        if cls._categorization_service is None:
+            cls._categorization_service = CategorizationService()
+        return cls._categorization_service

spendoo/ocr/pipeline.py

@@ -1,13 +1,12 @@
-from spendoo.ocr.service import OCRService
-from spendoo.categorization.service import CategorizationService
+from spendoo.core.services import ServiceContainer
 import base64
 
 class ReceiptPipeline:
 
     def __init__(self):
 
-        self.ocr = OCRService()
-        self.extractor = CategorizationService()
+        self.ocr = ServiceContainer.get_ocr_service()
+        self.extractor = ServiceContainer.get_categorization_service()
 
     def process_receipt(self, image_bytes):
 

spendoo/ocr/routes.py

@@ -1,5 +1,4 @@
 from fastapi import APIRouter, UploadFile, HTTPException
-import base64
 from spendoo.ocr.pipeline import ReceiptPipeline
 
 router = APIRouter(prefix="/ocr", tags=["OCR"])

spendoo/voice/routes.py

@@ -1,13 +1,28 @@
-from fastapi import APIRouter
+from fastapi import APIRouter, UploadFile, HTTPException
+from spendoo.core.services import ServiceContainer
+import os
+from spendoo.core.config import settings
 
-router = APIRouter(prefix="/voice", tags=["voice"])
+router = APIRouter(prefix="/voice", tags=["Voice"])
 
+voice_service = ServiceContainer.get_voice_service()
+categorization_service = ServiceContainer.get_categorization_service()
 
-@router.get("/")
-def info():
-    return {"module": "voice", "status": "ok"}
+@router.post("/process")
+async def process_voice(file: UploadFile):
 
+    ext = os.path.splitext(file.filename)[1].lower()
 
-@router.get("/health")
-def health():
-    return {"status": "ok"}
+    if ext not in settings.VOICE_ALLOWED_EXTENSIONS:
+        raise HTTPException(400, "Unsupported audio format")
+
+    audio_bytes = await file.read()
+
+    if len(audio_bytes) > settings.VOICE_MAX_SIZE:
+        raise HTTPException(400, "Audio file too large (max 25MB)")
+
+    text = voice_service.transcribe(audio_bytes, file.filename)
+
+    result = categorization_service.extract(text)
+
+    return result

spendoo/voice/service.py

@@ -1,5 +1,26 @@
-# Voice service placeholder
+from groq import Groq
+from spendoo.core.config import settings
 
-def get_status():
-    return {'module': 'voice', 'status': 'ok'}
 
+class VoiceService:
+
+    def __init__(self):
+        self.client = Groq(api_key=settings.GROQ_API_KEY)
+
+    def transcribe(self, audio_bytes: bytes, filename: str) -> str:
+        """
+        Convert audio → text using Whisper
+        """
+
+        transcription = self.client.audio.transcriptions.create(
+            file=(filename, audio_bytes),
+            model="whisper-large-v3",
+
+            prompt="""
+            Specify context and highlight on different languages used
+            """,
+
+            temperature=0
+        )
+
+        return transcription.text