change model

app.py

@@ -4,13 +4,12 @@ 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_name = "Ellbendls/csm-1b-indonesian-fine-tuned"
 model = VitsModel.from_pretrained(model_name)
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 
@@ -41,24 +40,19 @@ async def root():
 @app.post("/tts")
 async def text_to_speech(payload: dict):
     text = payload.get("text", "")
-    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 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:
         audio_data = await f.read()
 
-    # Build JSON response with download link
     return {
         "script": text,
         "timestamps": generate_timestamps(text),

requirements.txt

@@ -5,4 +5,5 @@ pydantic
 transformers
 torch
 torchaudio
-soundfile
+soundfile
+sentencepiece