Fix GPU timeout: increase to 300s, lower default diffusion steps to 10

First call downloads all models (DiT, Whisper, BigVGAN, RMVPE, CAMPPlus)
which takes significant time. Increase GPU duration to 300s (ZeroGPU PRO max).
Lower default diffusion steps from 25 to 10 (matching official Seed-VC Space)
for faster processing. Allow up to 100 steps for quality.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

app.py

@@ -309,10 +309,10 @@ with gr.Blocks(
                         )
                         convert_diffusion = gr.Slider(
                             minimum=5,
-                            maximum=50,
-                            value=25,
+                            maximum=100,
+                            value=10,
                             step=5,
-                            label="Qualite (plus haut = meilleure qualite, plus lent)",
+                            label="Qualite (10=rapide, 25=bon, 50-100=meilleure qualite)",
                         )
                         convert_vocal_vol = gr.Slider(
                             minimum=0.0,

pipeline/inference.py

@@ -184,7 +184,7 @@ def _test_import(name, module_path, subattr=None):
         return False, "{}: {}".format(type(ie).__name__, ie)
 
 
-@spaces.GPU(duration=120)
+@spaces.GPU(duration=300)
 def convert_voice(
     audio_path,
     reference_path,