Update app.py

app.py

@@ -1,36 +1,38 @@
 import torch
 import numpy as np
 import gradio as gr
-from transformers import AutoTokenizer, AutoModelForSpeechSeq2Seq
+from transformers import pipeline
 import logging
 from scipy.io.wavfile import write
 import uuid
 import os
+import warnings
 
 # -----------------------------
-# Reduce Transformers warnings
+# SUPPRESS WARNINGS
 # -----------------------------
+warnings.filterwarnings("ignore", category=FutureWarning)
 logging.getLogger("transformers").setLevel(logging.ERROR)
 
 # -----------------------------
-# LOAD LOCAL MODEL AND TOKENIZER
+# DEVICE SETUP
 # -----------------------------
 device = 0 if torch.cuda.is_available() else -1
-model_dir = "./"  # مسیر محلی در Space، همان‌جایی که adapter_model.safetensors قرار دارد
 
-# بارگذاری tokenizer
-tokenizer = AutoTokenizer.from_pretrained(model_dir)
+# -----------------------------
+# PATH TO FINE-TUNED MODEL
+# -----------------------------
+model_dir = "./"  # مسیر فایل‌های fine-tuned در Space
 
-# بارگذاری مدل
-model = AutoModelForSpeechSeq2Seq.from_pretrained(
-    model_dir,
-    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
-    low_cpu_mem_usage=True,
+# -----------------------------
+# LOAD TTS PIPELINE
+# -----------------------------
+tts_pipe = pipeline(
+    task="text-to-speech",
+    model=model_dir,
+    device=device
 )
 
-if device >= 0:
-    model = model.to(f"cuda:{device}")
-
 # -----------------------------
 # INFERENCE FUNCTION
 # -----------------------------
@@ -38,17 +40,16 @@ def tts_generate(text):
     if not text.strip():
         return None
 
-    # تبدیل متن به توکن
-    inputs = tokenizer(text, return_tensors="pt").to(model.device)
+    # اجرای مدل TTS
+    output = tts_pipe(text)
 
-    # تولید صوت
-    with torch.no_grad():
-        audio_out = model.generate_speech(**inputs)
+    if "audio" not in output:
+        raise ValueError("TTS pipeline did not return audio")
 
-    audio = np.array(audio_out.cpu().numpy(), dtype=np.float32)
+    audio = np.array(output["audio"], dtype=np.float32)
 
     # sampling rate پیش‌فرض
-    sr = 22050
+    sr = output.get("sampling_rate", 22050)
 
     # تبدیل float32 به int16 برای scipy
     audio_int16 = (audio * 32767).astype(np.int16)
@@ -82,10 +83,13 @@ demo = gr.Interface(
         placeholder=SAMPLES[0],
     ),
     outputs=gr.Audio(type="filepath", label="Generated Audio"),
-    title="Custom TTS with Adapter Model",
+    title="Fine-tuned Orpheus-3B Expressive TTS",
     examples=[[s] for s in SAMPLES],
 )
 
+# -----------------------------
+# CLEAN RUN
+# -----------------------------
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch(ssr_mode=False)  # ssr_mode=False برای کاهش خطاهای asyncio