test_1

requirements.txt

@@ -7,4 +7,6 @@ pydantic>=1.10.0
 python-dotenv>=1.0.0
 gradio>=4.0.0
 requests>=2.31.0
-librosa==0.10.2.post1
+librosa==0.10.2.post1
+joblib==1.4.2
+numpy==2.0.2

tasks/audio.py

@@ -4,6 +4,9 @@ from datasets import load_dataset
 from sklearn.metrics import accuracy_score
 import random
 import os
+import numpy as np
+import librosa
+import joblib
 
 from .utils.evaluation import AudioEvaluationRequest
 from .utils.emissions import tracker, clean_emissions_data, get_space_info
@@ -15,7 +18,42 @@ router = APIRouter()
 
 DESCRIPTION = "Random Baseline"
 ROUTE = "/audio"
-
+def create_spec(dataset):
+    spectograms = []
+    y = []
+    
+    for d in enumerate(dataset):  
+        audio_sample = d["audio"]["array"] if d["audio"]["sampling_rate"] == 12000 else librosa.resample(
+                d["audio"]["array"], 
+                orig_sr= d["audio"]["sampling_rate"], 
+                target_sr=12000
+            )
+        if len(audio_sample) == 0:
+            continue     
+        if len(audio_sample) < 36000:
+            padding_needed = 36000 - len(audio_sample)
+            repeats = (padding_needed // len(audio_sample)) + 1
+            audio_sample = np.concatenate([audio_sample] + [audio_sample[:padding_needed]] * repeats)[:36000]
+        elif len(audio_sample) > 36000:
+            audio_sample = audio_sample[:36000]
+        
+        rms = np.sqrt(np.mean(np.square(audio_sample)))
+        scalar = 10 ** (-20 / 20) / (rms + 1e-8)
+        
+        mel = librosa.feature.melspectrogram(
+            y=audio_sample*scalar,
+            sr=12000,
+            n_fft=2048,
+            hop_length=512,
+            n_mels=32,
+            power=2.0,
+        )
+        mel_db = librosa.power_to_db(mel, ref=np.max)  
+        mel_db_normalized = (mel_db - mel_db.mean()) / (mel_db.std() + 1e-8)
+        spectograms.append(np.float16(mel_db_normalized).T.flatten())
+        y.append(d["label"])
+        
+    return np.stack(spectograms),y
 
 
 @router.post(ROUTE, tags=["Audio Task"],
@@ -48,6 +86,11 @@ async def evaluate_audio(request: AudioEvaluationRequest):
     tracker.start()
     tracker.start_task("inference")
     
+    test_spec, y_test = create_spec(test_dataset)
+    H = np.load("array_file.npy")
+    W_val = np.dot(test_spec, H)
+    model_filename = 'model.joblib'
+    model = joblib.load(model_filename)
     #--------------------------------------------------------------------------------------------
     # YOUR MODEL INFERENCE CODE HERE
     # 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.
@@ -55,8 +98,7 @@ async def evaluate_audio(request: AudioEvaluationRequest):
     
     # Make random predictions (placeholder for actual model inference)
     true_labels = test_dataset["label"]
-    predictions = [random.randint(0, 1) for _ in range(len(true_labels))]
-    
+    predictions = model.predict(W_val)
     #--------------------------------------------------------------------------------------------
     # YOUR MODEL INFERENCE STOPS HERE
     #--------------------------------------------------------------------------------------------