Move all sensitive data to .env: API keys, IPs, URLs. Add example.env and .gitignore

.gitignore

@@ -0,0 +1,4 @@
+.env
+__pycache__/
+*.pyc
+.dockerenv

app.py

@@ -3,6 +3,8 @@
 Smart STT Pipeline v3
 T-one (fast) + GigaAM (accurate) + Qwen3.5-4B (punctuation)
 Auto-selects best STT based on confidence/silence detection.
+
+Configuration: all sensitive data loaded from .env file.
 """
 
 import os
@@ -10,22 +12,25 @@ import tempfile
 import time
 import subprocess
 import re
-import json
 
+from dotenv import load_dotenv
 from fastapi import FastAPI, File, UploadFile, HTTPException
 from pydantic import BaseModel
 import requests
 
-# === Configuration ===
+# Load .env (skip if not found)
-TONE_URL = "http://localhost:8003/transcribe"
+load_dotenv()
-GIGAAM_URL = "http://localhost:8001/transcribe"
+
-QWEN_URL = "http://178.17.143.46:8080/v1/chat/completions"
+# === Configuration from .env ===
-QWEN_URL_FALLBACK = "http://192.168.10.101:8888/v1/chat/completions"
+TONE_URL = os.getenv("TONE_URL", "http://localhost:8003/transcribe")
-QWEN_API_KEY = "f9LHodD0cOIaku1zu7i7vh9KZsb5EelGK4K85k1tJWgzablcQrFw-QNIJaEHHwMb3LZ5k1nd-CKYiYY0F5Yw"
+GIGAAM_URL = os.getenv("GIGAAM_URL", "http://localhost:8001/transcribe")
-QWEN_MODEL = "Qwen3.5-4B-Q5_K_M.gguf"
+QWEN_URL = os.getenv("QWEN_URL", "http://178.17.143.46:8080/v1/chat/completions")
-QWEN_MODEL_FALLBACK = "Qwen3.5-4B-MTP-GGUF"
+QWEN_URL_FALLBACK = os.getenv("QWEN_URL_FALLBACK", "http://192.168.10.101:8888/v1/chat/completions")
-MIN_TEXT_LENGTH = 5  # Minimum chars to consider T-one result valid
+QWEN_API_KEY = os.getenv("QWEN_API_KEY", "")
-SHORT_AUDIO_THRESHOLD = 10  # Seconds: if audio < this, always use GigaAM
+QWEN_MODEL = os.getenv("QWEN_MODEL", "Qwen3.5-4B-Q5_K_M.gguf")
+QWEN_MODEL_FALLBACK = os.getenv("QWEN_MODEL_FALLBACK", "Qwen3.5-4B-MTP-GGUF")
+MIN_TEXT_LENGTH = int(os.getenv("MIN_TEXT_LENGTH", "5"))
+SHORT_AUDIO_THRESHOLD = int(os.getenv("SHORT_AUDIO_THRESHOLD", "10"))
 
 app = FastAPI(title="Smart STT Pipeline v3", version="3.0.0")
 
@@ -37,7 +42,7 @@ class STTResponse(BaseModel):
     stt_time_sec: float
     punct_time_sec: float
     server_used: str
-    stt_engine: str  # "t-one", "gigaam", or "both"
+    stt_engine: str  # "t-one" or "gigaam"
 
 
 def convert_to_wav(audio_bytes: bytes) -> str:
@@ -61,14 +66,14 @@ def transcribe_tone(wav_path: str) -> dict:
     data = resp.json()
     if "detail" in data:
         return {"text": "", "valid": False, "error": data["detail"]}
-    
+
     text = data.get("text", "[]")
     phrases = re.findall(r"TextPhrase\(text='([^']*)'", text)
     result_text = " ".join(phrases)
-    
+
     # Check if result is meaningful
     valid = len(result_text.strip()) >= MIN_TEXT_LENGTH
-    
+
     return {
         "text": result_text,
         "valid": valid,
@@ -111,25 +116,25 @@ def add_punctuation(text: str, server_url: str, api_key: str, model: str) -> str
 async def transcribe(file: UploadFile = File(...)):
     """Smart transcription: T-one first, fallback to GigaAM if needed"""
     audio_bytes = await file.read()
-    
+
     # Get audio duration from file
     ogg_path = tempfile.mktemp(suffix=".ogg")
     with open(ogg_path, "wb") as f:
         f.write(audio_bytes)
-    
+
     cmd = ["ffprobe", "-v", "error", "-show_entries", "format=duration", "-of", "default=noprint_wrappers=1:nokey=1", ogg_path]
     proc = subprocess.run(cmd, capture_output=True, text=True)
     duration = float(proc.stdout.strip())
-    
+
     # Convert to WAV for T-one
     wav_path = convert_to_wav(audio_bytes)
-    
+
     try:
         # Step 1: Choose STT engine based on duration
         if duration < SHORT_AUDIO_THRESHOLD:
             # Short audio: use GigaAM (more accurate)
             giga_result = transcribe_gigaam(ogg_path)
-            
+
             text = giga_result.get("original", "")
             stt_engine = "gigaam"
             stt_time = giga_result.get("stt_time_sec", 0)
@@ -138,7 +143,7 @@ async def transcribe(file: UploadFile = File(...)):
             stt_start = time.time()
             tone_result = transcribe_tone(wav_path)
             stt_time = time.time() - stt_start
-            
+
             if tone_result["valid"]:
                 text = tone_result["text"]
                 stt_engine = "t-one"
@@ -148,7 +153,7 @@ async def transcribe(file: UploadFile = File(...)):
                 text = giga_result.get("original", "")
                 stt_engine = "gigaam"
                 stt_time = giga_result.get("stt_time_sec", 0)
-        
+
         # Step 2: Punctuation
         punct_start = time.time()
         server_used = "178.17.143.46:8080"
@@ -158,7 +163,7 @@ async def transcribe(file: UploadFile = File(...)):
             server_used = "192.168.10.101:8888"
             punctuated = add_punctuation(text, QWEN_URL_FALLBACK, QWEN_API_KEY, QWEN_MODEL_FALLBACK)
         punct_time = time.time() - punct_start
-        
+
         return {
             "original": text,
             "punctuated": punctuated,

example.env

@@ -0,0 +1,16 @@
+# T-one STT Service
+TONE_URL=http://localhost:8003/transcribe
+
+# GigaAM STT Service
+GIGAAM_URL=http://localhost:8001/transcribe
+
+# Qwen3.5-4B Punctuation Service
+QWEN_URL=http://178.17.143.46:8080/v1/chat/completions
+QWEN_URL_FALLBACK=http://192.168.10.101:8888/v1/chat/completions
+QWEN_API_KEY=your_qwen_api_key_here
+QWEN_MODEL=Qwen3.5-4B-Q5_K_M.gguf
+QWEN_MODEL_FALLBACK=Qwen3.5-4B-MTP-GGUF
+
+# Smart Routing Configuration
+MIN_TEXT_LENGTH=5
+SHORT_AUDIO_THRESHOLD=10

requirements.txt

@@ -2,3 +2,4 @@ fastapi==0.115.0
 uvicorn==0.32.0
 python-multipart==0.0.12
 requests==2.34.2
+python-dotenv==1.0.1