change model

app.py

@@ -1,12 +1,19 @@
 from fastapi import FastAPI
-from gtts import gTTS
+from transformers import VitsModel, AutoTokenizer
+import torch
 import aiofiles
 import uvicorn
 import os
 import time
+import soundfile as sf
 
 app = FastAPI()
 
+# Load model once at startup
+model_name = "facebook/mms-tts-ind"
+model = VitsModel.from_pretrained(model_name)
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
 def generate_timestamps(script: str, wpm: int = 150):
     sentences = [s.strip() for s in script.replace("\n", " ").split(".") if s.strip()]
     timestamps = []
@@ -34,16 +41,18 @@ async def root():
 @app.post("/tts")
 async def text_to_speech(payload: dict):
     text = payload.get("text", "")
-    lang = payload.get("lang", "id")
-    slow = payload.get("slow", False)
+    lang = payload.get("lang", "id")  # Ignored here, since model is Indonesian only
+    slow = payload.get("slow", False) # Also ignored
 
     if not text:
         return {"error": "Text is required"}
 
-    # Save audio file
-    file_path = "output.mp3"
-    tts = gTTS(text=text, lang=lang, slow=slow)
-    tts.save(file_path)
+    # Save audio file as WAV
+    file_path = "output.wav"
+    inputs = tokenizer(text, return_tensors="pt")
+    with torch.no_grad():
+        output = model(**inputs).waveform
+    sf.write(file_path, output.squeeze().cpu().numpy(), model.config.sampling_rate)
 
     # Read audio for size
     async with aiofiles.open(file_path, mode="rb") as f:
@@ -62,7 +71,7 @@ async def download_file(filename: str):
     file_path = os.path.join(os.getcwd(), filename)
     if os.path.exists(file_path):
         from fastapi.responses import FileResponse
-        return FileResponse(file_path, media_type="audio/mpeg", filename=filename)
+        return FileResponse(file_path, media_type="audio/wav", filename=filename)
     return {"error": "File not found"}
 
 if __name__ == "__main__":