Fix Gradio 6 Audio/Blocks params, harden duration estimator

app.py

@@ -148,13 +148,16 @@ def _score_to_ui(notes: list[dict]) -> str:
 def _estimate_duration(task: str, *args, **kwargs) -> int:
     """Callable @spaces.GPU duration: reserve what this request needs, not a blanket worst case.
 
-    Receives the same inputs as the decorated function, keyword-first: task is
-    inserted by each handler so we don't depend on positional layout. Calibrated
-    on an RTX PRO 6000: ~18 s cold-start overhead + flow_steps × ~55 ms per
-    generated patch + decode time.
+    Receives the same inputs as the decorated function (positional layout of
+    the handler signature when ZeroGPU passes args positionally, keywords as
+    a fallback — both are accepted). Calibrated on an RTX PRO 6000: ~18 s
+    cold-start overhead + flow_steps × ~55 ms per generated patch + decode
+    time.
     """
-    text = kwargs.get("text") or kwargs.get("score_text") or ""
+    text = kwargs.get("text") or kwargs.get("score_text") or (args[0] if args else "") or ""
     flow_steps = kwargs.get("flow_steps")
+    if flow_steps is None and len(args) >= 5:
+        flow_steps = args[4]
     steps = int(flow_steps) if flow_steps else 10
     steps = max(steps, 2)
     if task == "svs" and isinstance(text, str) and "|" in text:
@@ -342,7 +345,7 @@ def svs_examples() -> list[list]:
     return rows
 
 
-with gr.Blocks(theme=gr.themes.Citrus(), css=CSS, title="SemBridge — Speech & Singing") as demo:
+with gr.Blocks(title="SemBridge — Speech & Singing") as demo:
     gr.HTML(
         "<div style='text-align:center'>"
         "<h1>🎤 SemBridge</h1>"
@@ -369,8 +372,7 @@ with gr.Blocks(theme=gr.themes.Citrus(), css=CSS, title="SemBridge — Speech &
                     lines=2, autofocus=True,
                 )
                 tts_run = gr.Button("Generate speech", variant="primary")
-                tts_out = gr.Audio(label="Generated speech", type="filepath",
-                                   show_download_button=True)
+                tts_out = gr.Audio(label="Generated speech", type="filepath")
                 tts_status = gr.Textbox(label="Info", show_label=False, interactive=False)
                 with gr.Accordion("Reference voice (required)", open=True):
                     tts_voice = gr.Dropdown(list(VOICE_BY_LABEL) + [CUSTOM],
@@ -403,8 +405,7 @@ with gr.Blocks(theme=gr.themes.Citrus(), css=CSS, title="SemBridge — Speech &
                     lines=8, autofocus=True,
                 )
                 svs_run = gr.Button("Generate singing", variant="primary")
-                svs_out = gr.Audio(label="Generated singing", type="filepath",
-                                   show_download_button=True)
+                svs_out = gr.Audio(label="Generated singing", type="filepath")
                 svs_status = gr.Textbox(label="Info", show_label=False, interactive=False)
                 with gr.Accordion("Sung reference voice (required)", open=True):
                     svs_voice = gr.Dropdown(list(VOICE_BY_LABEL) + [CUSTOM],
@@ -440,8 +441,7 @@ with gr.Blocks(theme=gr.themes.Citrus(), css=CSS, title="SemBridge — Speech &
                     label="Case (speech → singing → speech)",
                 )
                 uni_run = gr.Button("Generate mixed sequence", variant="primary")
-                uni_out = gr.Audio(label="Generated mixed audio", type="filepath",
-                                   show_download_button=True)
+                uni_out = gr.Audio(label="Generated mixed audio", type="filepath")
                 uni_status = gr.Textbox(label="Info", show_label=False, interactive=False)
                 with gr.Accordion("Generation overrides (optional)", open=False):
                     with gr.Row():
@@ -476,4 +476,4 @@ with gr.Blocks(theme=gr.themes.Citrus(), css=CSS, title="SemBridge — Speech &
                   inputs=[case_id, uni_flow, uni_cfg, uni_temp, uni_seed],
                   outputs=[uni_out, uni_status], api_name="unified")
 
-demo.launch(mcp_server=True)
+demo.launch(mcp_server=True, theme=gr.themes.Citrus(), css=CSS)