make audioset the default weights, otherwise no weights are loaded

[hutauf]

``If I'm using the demo code from the readme, the models are initiated with random weights and the audioset-weights are not being loaded. Here is some demo-code showing that the model is different. This should give two identical results. But if you remove the "weights" argument, it will give two different results.

import librosa
import numpy as np
import vggish_keras as vgk
import tensorflow as tf

y, sr = librosa.load(librosa.util.example_audio_file())

tf.random.set_seed(1)
compute = vgk.get_embedding_function(hop_duration=0.25, weights='audioset')
Z, ts = compute(y=y, sr=sr)
print("first run", Z.flatten()[:10])

tf.random.set_seed(2)
compute = vgk.get_embedding_function(hop_duration=0.25, weights='audioset')
Z, ts = compute(y=y, sr=sr)
print("second run", Z.flatten()[:10])