predict.py

import numpy as np
import tensorflow
import librosa 
import numpy as np 
import pickle

# Change the path
model = tensorflow.keras.models.load_model('weights.best.basic_mlp.hdf5', compile=False)      


def extract_feature(file_name):
   
    try:
        audio_data, sample_rate = librosa.load(file_name) 
        mfccs = librosa.feature.mfcc(y=audio_data, sr=sample_rate, n_mfcc=40)
        mfccsscaled = np.mean(mfccs.T,axis=0)
        
    except Exception as e:
        print("Error encountered while parsing file: ")
        return None, None

    return np.array([mfccsscaled])


def print_prediction(file_name):
    prediction_feature = extract_feature(file_name) 

    predicted_vector = model.predict(prediction_feature)
    array=predicted_vector[0]
    m=np.amax(array)
    if(m==array[3]):
      print("Accuracy:",array[3])
      print("Dog Detected")
    else:
      print("Accuracy:",1-array[3])
      print("No Dog Detected")
    
filename = input("Enter full path of audio(.wav) file: ") 
print_prediction(filename)