Update app.py

app.py

@@ -1,22 +1,32 @@
+import os
 import gradio as gr
-from transformers import pipeline
-import scipy
+from transformers import BarkModel, AutoProcessor
+import torch
 
-def getText(text):
-    return text
+# Desactivar paralelismo en tokenizers para evitar advertencias
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
 def say_something(text):
-    synthesiser = pipeline("text-to-speech", "suno/bark-small")
-    speech = synthesiser("Hello world!", forward_params={"do_sample": True})
-    scipy.io.wavfile.write("bark_out.wav", rate=speech["sampling_rate"], data=speech["audio"])
-    return "bark_out.wav"
+    processor = AutoProcessor.from_pretrained("suno/bark-small")
+
+    model = BarkModel.from_pretrained("suno/bark-small").to(device)
+
+    inputs = processor(text)
+    audio_array = model.generate(**inputs)
+    audio_array = audio_array.cpu().numpy().squeeze()
+
+    # Devuelve el audio como un numpy array junto con el sample rate
+    sample_rate = model.generation_config.sample_rate
+    return (sample_rate, audio_array)
     
 
 with gr.Blocks() as demo:
-    textBox = gr.Textbox(label="Text")
+    textBox = gr.Textbox(label="Text", value="Hello! [laugths]. This is a test!")
     button = gr.Button("Generate Speech")
     audio_output = gr.Audio(label="Generated Speech")
-    button.click( say_something, outputs=[audio_output])
+    button.click(say_something, inputs=textBox, outputs=audio_output)
 
 
 demo.launch()