Update app.py

app.py

@@ -70,6 +70,51 @@ def predict(filepath):
     tensor = index_to_label(tensor.squeeze())
     return tensor
     
-model = torch.load('export.pkl',map_location=torch.device('cpu'))
+def record(seconds=1):
+
+    from google.colab import output as colab_output
+    from base64 import b64decode
+    from io import BytesIO
+    from pydub import AudioSegment
+
+    RECORD = (
+        b"const sleep  = time => new Promise(resolve => setTimeout(resolve, time))\n"
+        b"const b2text = blob => new Promise(resolve => {\n"
+        b"  const reader = new FileReader()\n"
+        b"  reader.onloadend = e => resolve(e.srcElement.result)\n"
+        b"  reader.readAsDataURL(blob)\n"
+        b"})\n"
+        b"var record = time => new Promise(async resolve => {\n"
+        b"  stream = await navigator.mediaDevices.getUserMedia({ audio: true })\n"
+        b"  recorder = new MediaRecorder(stream)\n"
+        b"  chunks = []\n"
+        b"  recorder.ondataavailable = e => chunks.push(e.data)\n"
+        b"  recorder.start()\n"
+        b"  await sleep(time)\n"
+        b"  recorder.onstop = async ()=>{\n"
+        b"    blob = new Blob(chunks)\n"
+        b"    text = await b2text(blob)\n"
+        b"    resolve(text)\n"
+        b"  }\n"
+        b"  recorder.stop()\n"
+        b"})"
+    )
+    RECORD = RECORD.decode("ascii")
+
+    print(f"Recording started for {seconds} seconds.")
+    display(ipd.Javascript(RECORD))
+    s = colab_output.eval_js("record(%d)" % (seconds * 1000))
+    print("Recording ended.")
+    b = b64decode(s.split(",")[1])
+
+    fileformat = "wav"
+    filename = f"_audio.{fileformat}"
+    AudioSegment.from_file(BytesIO(b)).export(filename, format=fileformat)
+    return torchaudio.load(filename)
+
     
-gr.Interface(fn=predict, inputs=gr.inputs.Audio(source='microphone',type='filepath'), outputs=gr.outputs.Label(num_top_classes=3)).launch(share=True)
+model = torch.load('export.pkl',map_location=torch.device('cpu'))
+
+  
+      
+gr.Interface(fn=predict, inputs=gr.inputs.Audio(source=record()[0]), outputs=gr.outputs.Label(num_top_classes=3)).launch(share=True)