Update app.py

app.py

@@ -1,20 +1,18 @@
-import os
+import torch
+from transformers import pipeline
 import gradio as gr
-from huggingface_hub import InferenceClient
 import numpy as np
-import io
 import soundfile as sf
+import io
 
 # -----------------------------
-# HUGGING FACE INFERENCE CLIENT
+# LOAD PIPELINE
 # -----------------------------
-HF_TOKEN = os.getenv("HF_TOKEN")
-if HF_TOKEN is None:
-    raise ValueError("Environment variable HF_TOKEN not found.")
-
-client = InferenceClient(
-    provider="auto",  # می‌توانید 'fal-ai' یا 'auto' استفاده کنید
-    api_key=HF_TOKEN,
+device = 0 if torch.cuda.is_available() else -1
+tts_pipe = pipeline(
+    task="text-to-speech",
+    model="canopylabs/orpheus-3b-0.1-ft",
+    device=device
 )
 
 # -----------------------------
@@ -25,24 +23,25 @@ def tts_generate(text):
         return None
 
     try:
-        # دریافت صوت به صورت bytes
-        audio_bytes = client.text_to_speech(
-            text,
-            model="canopylabs/orpheus-3b-0.1-ft",
-        )
+        output = tts_pipe(text)
     except Exception as e:
-        return f"Error: {str(e)}"
+        print("Error:", e)
+        return None
+
+    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")
+    # Convert to bytes for Gradio Audio
+    buffer = io.BytesIO()
+    sf.write(buffer, audio, sr, format="WAV")
+    buffer.seek(0)
 
-    return (sr, audio)
+    return buffer
 
 # -----------------------------
-# ENGLISH SAMPLE TEXTS
+# SAMPLE TEXTS WITH STYLE TAGS
 # -----------------------------
-ENGLISH_SAMPLES = [
+SAMPLES = [
     "[neutral] Hello! This is a neutral English voice generated by Orpheus.",
     "[expressive] I'm really excited to show you how natural this voice sounds!",
     "[calm] Please relax and enjoy this calm and smooth narration.",
@@ -57,13 +56,13 @@ demo = gr.Interface(
     fn=tts_generate,
     inputs=gr.Textbox(
         label="Enter text (English supported with tags)",
-        placeholder=ENGLISH_SAMPLES[0],
-        lines=4,
+        placeholder=SAMPLES[0],
+        lines=4
     ),
-    outputs=gr.Audio(label="Generated Audio"),
-    title="Orpheus 3B Text-to-Speech (Inference API)",
+    outputs=gr.Audio(type="file", label="Generated Audio"),
+    title="Orpheus‑3B TTS",
     description=(
-        "English TTS using **canopylabs/orpheus-3b-0.1-ft** via Hugging Face Inference API.\n\n"
+        "English TTS using **canopylabs/orpheus-3b-0.1-ft** via Transformers pipeline.\n\n"
         "Supported style tags examples:\n"
         "- `[neutral]`\n"
         "- `[expressive]`\n"
@@ -71,9 +70,9 @@ demo = gr.Interface(
         "- `[narration]`\n"
         "- `[conversation]`\n\n"
         "Example:\n"
-        "`[expressive] I'm very happy to see you today!.`"
+        "`[expressive] I'm very happy to see you today!`"
     ),
-    examples=[[s] for s in ENGLISH_SAMPLES],
+    examples=[[s] for s in SAMPLES],
 )
 
 if __name__ == "__main__":