add app

app.py

@@ -1,17 +1,31 @@
 import gradio as gr
-from transformers import pipeline
+from transformers import pipeline, AutoProcessor
 import numpy as np
 
-# Initialize the model
+# Initialize the model and processor
+processor = AutoProcessor.from_pretrained("suno/bark-small")
 pipe = pipeline(model="suno/bark-small")
 
 def text_to_speech(text):
+    # Prepare the input with proper attention mask
+    inputs = processor(
+        text,
+        return_tensors="pt",
+        padding=True,
+        return_attention_mask=True
+    )
+    
     # Generate audio from text
-    output = pipe(text)
+    output = pipe(
+        text,
+        attention_mask=inputs.attention_mask
+    )
     
-    # Normalize audio to prevent clipping
+    # Normalize and scale audio to int16 range
     audio = output["audio"]
+    audio = np.float32(audio)  # Ensure float32 type
     audio = audio / np.max(np.abs(audio))  # Normalize to [-1, 1]
+    audio = (audio * 32767).astype(np.int16)  # Convert to int16 range
     
     return (output["sampling_rate"], audio)
 
@@ -22,7 +36,7 @@ demo = gr.Interface(
         label="Text to speak",
         placeholder="Enter the text you want to convert to speech...",
     ),
-    outputs=gr.Audio(label="Generated Speech", type="numpy"),  # Specify numpy type
+    outputs=gr.Audio(label="Generated Speech"),
     title="Text to Speech with Bark-small",
     description="Convert text to speech using the Suno Bark-small model",
     examples=[

requirements.txt

@@ -2,4 +2,5 @@ gradio
 transformers
 torch
 accelerate
-numpy
+numpy
+scipy