I think the key thing has been enough processing for dynamic/real time spatial audio via HRTF. Despite being “just” audio, applying a HRTF can be pretty processing intensive, especially for a mobile device. Those binaural audio experiences were all pre-computed.