Why is Tensorflow.js fftSize different by a factor of two to the Web API fftSize?

Question

tf.data.microphone has an fftSize parameter, described as "the number of samples used to compute each nonoverlapping “frame” of audio" in the Deep Lea