ZeroGPU

app.py

@@ -1,6 +1,7 @@
 import sys
 import os
 import torch
+import spaces
 # By using XTTS you agree to CPML license https://coqui.ai/cpml
 os.environ["COQUI_TOS_AGREED"] = "1"
 
@@ -50,7 +51,7 @@ def predict(prompt, language, audio_file_pth, mic_file_path, use_mic):
                 None,
             )
     if len(prompt) > 50000:
-        gr.Warning("Text length limited to 50000 characters for this demo, please try shorter text")
+        gr.Warning("Text length limited to 50,000 characters for this demo, please try shorter text")
         return (
                 None,
                 None,
@@ -62,12 +63,7 @@ def predict(prompt, language, audio_file_pth, mic_file_path, use_mic):
                 language = "fr-fr"
         if m.find("/fr/") != -1:
             language = None
-        tts.tts_to_file(
-            text=prompt,
-            file_path="output.wav",
-            speaker_wav=speaker_wav,
-            language=language
-        )
+        predict_on_gpu(prompt, peaker_wav, language)
     except RuntimeError as e :
         if "device-assert" in str(e):
             # cannot do anything on cuda device side error, need to restart
@@ -85,6 +81,15 @@ def predict(prompt, language, audio_file_pth, mic_file_path, use_mic):
         None,
     )
 
+@spaces.GPU(duration=60)
+def predict_on_gpu(prompt, peaker_wav, language):
+    tts.tts_to_file(
+        text=prompt,
+        file_path="output.wav",
+        speaker_wav=speaker_wav,
+        language=language
+    )
+
 with gr.Blocks() as interface:
     gr.HTML("Multi-language Text-to-Speech")
     gr.HTML(