Update app.py

app.py

@@ -5,16 +5,20 @@ import numpy as np
 import matplotlib.pyplot as plt
 from pydub import AudioSegment
 import tempfile
+import PIL
 
 learn = load_learner('model.pkl')
 labels = learn.dls.vocab
 
 def audio_to_spectrogram(audio_file):
-    if audio_file.endswith('.mp3'):
-        with tempfile.NamedTemporaryFile(suffix='.wav') as temp_wav:
-            audio = AudioSegment.from_mp3(audio_file)
-            audio.export(temp_wav.name, format='wav')
-            y, sr = librosa.load(temp_wav.name, sr=None)
+    if isinstance(audio_file, str):
+        if audio_file.endswith('.mp3'):
+            with tempfile.NamedTemporaryFile(suffix='.wav') as temp_wav:
+                audio = AudioSegment.from_mp3(audio_file)
+                audio.export(temp_wav.name, format='wav')
+                y, sr = librosa.load(temp_wav.name, sr=None)
+        else:
+            y, sr = librosa.load(audio_file, sr=None)
     else:
         y, sr = librosa.load(audio_file, sr=None)
 
@@ -36,11 +40,11 @@ def predict(audio):
     pred, pred_idx, probs = learn.predict(img)
     return {labels[i]: float(probs[i]) for i in range(len(labels))}
 
-examples = ['example_audio.mp3']
-
 gr.Interface(
     fn=predict,
-    inputs=gr.Audio(sources="upload", type="filepath", label="Upload audio (WAV or MP3)"),
+    inputs=[
+        gr.Audio(sources=["upload", "microphone"], type="filepath", label="Upload or Record audio (WAV or MP3)"),
+    ],
     outputs=gr.components.Label(num_top_classes=3),
-    examples=examples,
-).launch()
+    live=True
+).launch()