I agree that "intermittent" is interesting. However, there was time for two other warnings to be recorded after that, so it implies the speed was falling earlier rather than later in the short final sequence.
The report doesn't state that there was a failure to enter autorotation. The lack of the SHED BUS switching implies it never became stable or relaxed enough to permit that to be done.
There's a few things that might illuminate that scenario -
The physical damage to the wreckage will eventually reveal a vertical speed on impact, but looks to be quite fast. The term "very high energy deceleration" is used in the report.
The helicopter landed with mostly vertical speed on a roof. That seems to be inconsistent with stretching out distance over the river. The radar track prior to the accident should make this clear.
If the 400ft radar return is from a single reading on descent, after 1000ft and before zero, then that sets a limit to the descent speed of something more than about 50ft/s, assuming a 10s scan. If some of that was more gentle, then the end has to be steeper.