add /understand API endpoint for GGUF audio captioning

app.py

@@ -289,6 +289,38 @@ def _start_ace_server():
     return False
 
 
+# ---------------------------------------------------------------------------
+# /understand proxy -- caption audio via ace-server GGUF
+# ---------------------------------------------------------------------------
+
+def understand_audio(audio_file):
+    """Send audio to ace-server /understand endpoint, return metadata."""
+    if not _server_ok():
+        return "ace-server not running"
+    if audio_file is None:
+        return "No audio file"
+    src = audio_file.name if hasattr(audio_file, "name") else str(audio_file)
+    import base64
+    with open(src, "rb") as f:
+        audio_b64 = base64.b64encode(f.read()).decode()
+    try:
+        r = requests.post(
+            f"{ACE_SERVER}/understand",
+            json={"audio": audio_b64},
+            timeout=300,
+        )
+        if r.status_code != 200:
+            return f"Error {r.status_code}: {r.text}"
+        job_id = r.json().get("id")
+        status, elapsed = _poll_job(job_id, timeout=600)
+        if status != "done":
+            return f"Understand {status} after {elapsed:.0f}s"
+        result = _fetch_result(job_id)
+        return json.dumps(result.json(), indent=2)
+    except Exception as e:
+        return f"Error: {e}"
+
+
 # ---------------------------------------------------------------------------
 # CLI mode
 # ---------------------------------------------------------------------------
@@ -769,6 +801,17 @@ def gradio_main():
                     api_name="check_log",
                 )
 
+        # Hidden API endpoint for audio understanding (no UI, API-only)
+        understand_btn = gr.Button(visible=False)
+        understand_input = gr.File(visible=False)
+        understand_output = gr.Textbox(visible=False)
+        understand_btn.click(
+            understand_audio,
+            inputs=[understand_input],
+            outputs=[understand_output],
+            api_name="understand",
+        )
+
         demo.launch(
             server_name="0.0.0.0",
             server_port=7860,