To be more specific: starting listening and calling suspend is a workaround that developers came up with in order to hide the calibration time from users. It was a bad workaround because it resulted in the calibration being done against a different environment than the speech which arrived at some arbitrary later point, so you can find a few different 1.x threads here in which I recommend that developers not do it. Now there is no longer any reason for the workaround, since there is no calibration time when you start up, so it’s a good time to let go of the workaround architecture in which startListening is called at a time in which you don’t want to start listening.