Create app.py

app.py

@@ -0,0 +1,44 @@
+from transformers import pipeline
+import torch
+
+device = "cuda:0" if torch.cuda.is_available() else "cpu"
+
+classifier = pipeline(
+    "audio-classification", model="MIT/ast-finetuned-speech-commands-v2", device=device
+)
+
+from transformers.pipelines.audio_utils import ffmpeg_microphone_live
+
+def launch_fn(
+    wake_word = "sheila",
+    prob_threshold=0.5,
+    chunk_length_s=20.0,
+    stream_chunk_s=0.25,
+    debug=True
+    ):
+
+  print("called")
+  if wake_word not in list(classifier.model.config.id2label.values()):
+    raise ValueError(f"wake word must be one of {list(classifier.model.config.id2label.values())}")
+
+    sampling_rate = classifier.feature_extractor.sampling_rate
+
+    mic = ffmpeg_microphone_live(
+        sampling_rate=sampling_rate,
+        chunk_length_s=chunk_length_s,
+        stream_chunk_s=stream_chunk_s
+    )
+
+    print("listening for wake up word.......")
+
+    for prediction in classifier(mic):
+      prediction = prediction[0]
+      if debug == True:
+        print(prediction)
+      if (prediction['label'] == wake_word) and (prediction['score']  > prob_threshold):
+        return True, f"Wake word '{wake_word}' detected with probability {prediction['score']:.2f}"
+
+    return False, "Wake word not detected"
+
+if __name__ == "__main__":
+    launch_fn(debug=True)