Could the Omicron analysis windows length not be restricted to be a power of 2 anymore?
To test for instance a 6-s analysis window, in between the 4-s and 8-s analysis windows that are currently possible, while keeping an overlap of 4 seconds. That would mean decreasing the latency from 8-4/2 = 6s to 6-4/2 = 4s, while doubling the number of windows to be analyzed (whose unit size will be reduced from 8 to 6 seconds).