Update app.py

app.py

@@ -18,18 +18,19 @@ asr = pipeline(
     generate_kwargs={"language": "en", "task": "transcribe"}
 )
 
-def classify(audio):
+def classify(audio_path):
     """
-    audio: Tuple(numpy_array, sampling_rate)
+    audio_path: str → path to uploaded/recorded file
     returns: transcript (str), safety probabilities (dict), unsafe probability (str)
     """
-    arr, sr = audio
-    arr = arr.astype("float32")
-    if arr.ndim > 1:
-        arr = arr.mean(axis=1)
-
-    # ASR transcription
-    result = asr({"array": arr, "sampling_rate": sr})
+    # Read file from disk
+    audio, sr = sf.read(audio_path, dtype="float32")
+    # If stereo, convert to mono
+    if audio.ndim > 1:
+        audio = audio.mean(axis=1)
+
+    # ASR
+    result = asr({"array": audio, "sampling_rate": sr})
     txt = result["text"].strip()
 
     # Safety classification
@@ -38,16 +39,16 @@ def classify(audio):
     unsafe_str = f"{proba:.2f}"
     return txt, label_probs, unsafe_str
 
-# Use the components API
-audio_input      = gr.components.Audio(label="Upload or record audio", type="numpy")
-transcript_out   = gr.components.Textbox(label="Transcript")
-probs_out        = gr.components.Label(num_top_classes=2, label="Safety Probabilities")
-unsafe_prob_out  = gr.components.Textbox(label="Unsafe Probability")
+# Use filepath-based Audio component
+audio_input     = gr.components.Audio(label="Upload or record audio", type="filepath")
+transcript_out  = gr.components.Textbox(label="Transcript")
+probs_out       = gr.components.Label(num_top_classes=2, label="Safety Probabilities")
+unsafe_out      = gr.components.Textbox(label="Unsafe Probability")
 
 iface = gr.Interface(
     fn=classify,
     inputs=audio_input,
-    outputs=[transcript_out, probs_out, unsafe_prob_out],
+    outputs=[transcript_out, probs_out, unsafe_out],
     title="BubbleGuard Audio Safety Checker",
     description="Upload or record audio; get ASR transcript plus safe/unsafe probabilities."
 )