def Predict(
self,
model, # can be the model itself or model path.
spectrograms, # spectrograms that are generated from the given track.
output_directory=None,
track_name=None):
''' this function takes a series of spectrograms and then predict them in parallel then returns a concatenated spectrograms'''
if type(model) == str: # if model is a path not real object.
if not os.path.exists(model):
print('Model does not exist')
return None
model = tf.keras.models.load_model(model)
# Prediction part.----------------------------------------------
start_time = time.time()
manager = Pool()
outputs = manager.go(model.predict, spectrograms)
outputs = [outputs[i][0] for i in range(len(spectrograms))]
full_output_stem = reduce(lambda a, b: np.concatenate([a, b], axis=0),
outputs)
prediction_time = time.time() - start_time
print(
f'[LOG] prediction done. Prediction time: {prediction_time:.3f} seconds'
)
return full_output_stem
