Update app.py

app.py

@@ -1,33 +1,20 @@
-import torch
-import gradio as gr
-from transformers import pipeline
-from huggingface_hub import login
 import os
+import gradio as gr
+from huggingface_hub import InferenceClient
 import numpy as np
+import io
+import soundfile as sf
 
 # -----------------------------
-# HUGGING FACE LOGIN
+# HUGGING FACE INFERENCE CLIENT
 # -----------------------------
-HF_TOKEN = os.getenv(HF_TOKEN)
+HF_TOKEN = os.getenv("HF_TOKEN")
 if HF_TOKEN is None:
-    raise ValueError("Environment variable HF_TOKEN not found. Please set it before running.")
-
-login(HF_TOKEN)
-
-# -----------------------------
-# CONFIG
-# -----------------------------
-MODEL_NAME = "canopylabs/orpheus-3b-0.1-ft"
-DEVICE = 0 if torch.cuda.is_available() else -1
+    raise ValueError("Environment variable HF_TOKEN not found.")
 
-# -----------------------------
-# LOAD PIPELINE
-# -----------------------------
-tts_pipe = pipeline(
-    task="text-to-speech",
-    model=MODEL_NAME,
-    device=DEVICE,
-    use_auth_token=HF_TOKEN  # دسترسی به مدل gated
+client = InferenceClient(
+    provider="fal-ai",
+    api_key=HF_TOKEN,
 )
 
 # -----------------------------
@@ -37,10 +24,15 @@ def tts_generate(text):
     if not text.strip():
         return None
 
-    output = tts_pipe(text)
+    # دریافت صوت به صورت bytes
+    audio_bytes = client.text_to_speech(
+        text,
+        model="canopylabs/orpheus-3b-0.1-ft",
+    )
 
-    audio = np.asarray(output["audio"], dtype=np.float32)
-    sr = output["sampling_rate"]
+    # تبدیل bytes به numpy array و نمونه‌برداری
+    buffer = io.BytesIO(audio_bytes)
+    audio, sr = sf.read(buffer, dtype="float32")
 
     return (sr, audio)
 
@@ -66,9 +58,9 @@ demo = gr.Interface(
         lines=4,
     ),
     outputs=gr.Audio(label="Generated Audio"),
-    title="Orpheus 3B Text-to-Speech",
+    title="Orpheus 3B Text-to-Speech (Inference API)",
     description=(
-        "English TTS using **canopylabs/orpheus-3b-0.1-ft**\n\n"
+        "English TTS using **canopylabs/orpheus-3b-0.1-ft** via Hugging Face Inference API.\n\n"
         "Supported style tags examples:\n"
         "- `[neutral]`\n"
         "- `[expressive]`\n"