Fix(app): init ASR model in background; add error handling; remove share=True

app.py

@@ -1,4 +1,6 @@
 import time
+import threading
+import os
 
 import gradio as gr
 import librosa
@@ -10,7 +12,7 @@ AUDIO_SECONDS_THRESHOLD = 2
 #pipe = pipeline("audio-classification", model="MIT/ast-finetuned-audioset-10-10-0.4593")
 prediction = [{"score": 1, "label": "recording..."}]
 
-from server_wrapper import process_chunk_from_bytes
+from server_wrapper import process_chunk_from_bytes, init_model
 
 
 
@@ -27,7 +29,13 @@ def streaming_recording_fn(stream, new_chunk):
     y = librosa.resample(y, orig_sr=sr, target_sr=TARGET_SAMPLE_RATE)
     if stream is not None:
         if (stream.shape[-1] / TARGET_SAMPLE_RATE) >= AUDIO_SECONDS_THRESHOLD:
-            prediction = process_chunk_from_bytes((stream * 32768).astype(np.int16).tobytes())
+            try:
+                prediction = process_chunk_from_bytes((stream * 32768).astype(np.int16).tobytes())
+            except Exception as e:
+                print(f"process_chunk_from_bytes failed: {e}")
+                prediction = [{"score": 0.0, "label": "error"}]
+
+            os.makedirs('./audio', exist_ok=True)
             file_name = f'./audio/{time.strftime("%Y%m%d_%H%M%S", time.localtime())}.wav'
             # # sf.write(file_name, stream, TARGET_SAMPLE_RATE)
             print(f"SAVE AUDIO: {file_name}")
@@ -58,5 +66,13 @@ with gr.Blocks() as demo:
                        ["Streaming"])
 
 if __name__ == "__main__":
-
-    demo.launch(share=True)
+    # Initialize the ASR model in a background thread to avoid blocking the UI startup.
+    try:
+        t = threading.Thread(target=init_model, daemon=True)
+        t.start()
+        print("Background model initialization started")
+    except Exception as e:
+        print(f"Failed to start background model init: {e}")
+
+    # Do not use `share=True` inside Spaces; launch normally.
+    demo.launch()