This is what I was thinking. I might guess (wildly) that the noise cancelling function results in a tiny delay, leading to a DAF-like result. I'd be interested to find out if this is a common result...