Tweet
Each of the eight analysis bars provides a probability estimate that a negative factor is present, as evidenced by a colony's sound profile. Each is scaled from 0 % (no probability) to 100 % (100% or nearly 100% certainty). For factors such as Queenright or Queenless, where the analysis answer is Yes or No, a probability of 50% would essentially be the flip of a coin as to the likelihood that there isn't a laying queen. The first step here would be to re-record for 60 seconds. If the probability drops, the colony is likely to have a queen, but maybe she's not laying. If it comes back at 80%, one would really want to check by visual inspection.
For factors that vary in degree of infection or infestation, we expect to see a stronger % probability as the severity of the problem increases.
However, initially, given differences in audio hardware in smartphones, regional dialects of bees, we caution that the aforementioned interpretations will only apply once the app has been rigorously tuned. And for those factors that vary in severity, we need to obtain lots of recordings so that we can identify discrimination thresholds.
Finally, there's the explanation of Not Normal. Not Normal means that the pattern-matching AIs were unable to identify audio patterns indicative of healthy colonies. That could be a dead-out, a failing colony, or maybe even a pest or disease for which the app has yet to be calibrated. As a first step, we want to be able to break-out the dead-out colonies, rather than have the app report these as not normal. Real dead-outs are useful, but everyone can set up simulated dead out colonies with and without combs of honey.
For factors that vary in degree of infection or infestation, we expect to see a stronger % probability as the severity of the problem increases.
However, initially, given differences in audio hardware in smartphones, regional dialects of bees, we caution that the aforementioned interpretations will only apply once the app has been rigorously tuned. And for those factors that vary in severity, we need to obtain lots of recordings so that we can identify discrimination thresholds.
Finally, there's the explanation of Not Normal. Not Normal means that the pattern-matching AIs were unable to identify audio patterns indicative of healthy colonies. That could be a dead-out, a failing colony, or maybe even a pest or disease for which the app has yet to be calibrated. As a first step, we want to be able to break-out the dead-out colonies, rather than have the app report these as not normal. Real dead-outs are useful, but everyone can set up simulated dead out colonies with and without combs of honey.