Update app.py

app.py

@@ -1,5 +1,7 @@
 print("import gradio")
 import gradio as gr
+from scipy.io.wavfile import write
+import tempfile
 print("import ppn")
 from pypinyin import lazy_pinyin, Style
 print("import torch")
@@ -80,7 +82,9 @@ def speak(text):
     diffusion_conditioning = normalize_tacotron_mel(cond_mel)
     mel = do_spectrogram_diffusion(diffusion, diffuser, latent, diffusion_conditioning, temperature=1.0).detach().cpu()
     wav = vocos.decode(mel)
-    return (24000, wav.detach().cpu())
+    with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as f:
+        write(f.name, data=wav, rate=24000)
+        return f.name
 
 with gr.Blocks() as demo:
     gr.Markdown('# TTTS\n\nAn **unofficial** demo of [TTTS](https://github.com/adelacvg/ttts) based on XTTS. TTTS only supports Chinese.')