init

tasks/audio.py

@@ -1,3 +1,7 @@
+import librosa
+import joblib
+import numpy as np
+
 from fastapi import APIRouter
 from datetime import datetime
 from datasets import load_dataset
@@ -53,9 +57,21 @@ async def evaluate_audio(request: AudioEvaluationRequest):
     # Update the code below to replace the random baseline by your model inference within the inference pass where the energy consumption and emissions are tracked.
     #--------------------------------------------------------------------------------------------   
     
-    # Make random predictions (placeholder for actual model inference)
-    true_labels = test_dataset["label"]
-    predictions = [random.randint(0, 1) for _ in range(len(true_labels))]
+    def extract_features(example, sampling_rate):
+        audio_array = example['audio']['array']
+        mfcc = librosa.feature.mfcc(y=audio_array, sr=sampling_rate, n_mfcc=5)
+        return np.mean(mfcc, axis=1)
+
+    def predict_new_audio(model, dataset, sampling_rate):
+        features_list = [extract_features(example, sampling_rate) for example in dataset]
+        features_array = np.vstack(features_list)
+        predictions = model.predict(features_array)
+        return predictions
+
+    model_filename = "models/lightgbm_baseline_87_acc.pkl"
+    clf = joblib.load(model_filename)
+
+    predictions = predict_new_audio(clf, test_dataset, 12000)
     
     #--------------------------------------------------------------------------------------------
     # YOUR MODEL INFERENCE STOPS HERE
@@ -65,6 +81,7 @@ async def evaluate_audio(request: AudioEvaluationRequest):
     emissions_data = tracker.stop_task()
     
     # Calculate accuracy
+    true_labels = test_dataset["label"]
     accuracy = accuracy_score(true_labels, predictions)
     
     # Prepare results dictionary

tasks/models/lightgbm_baseline_87_acc.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f6ffd6404837979e258257c20667274785d1803ee107dc82967ce9d41cd4ced
+size 3388964