Date: 3/25/26 10:41 am From: cossypha via groups.io <leonard.janetl...> Subject: Re: [pen-bird] Merlin Audio recordings - Rufous and Allen's Hummingbirds - please read.
Most of these audio id programs use machine learning algorithms for ID;
they do not know the full range of any species' vocalizations let alone any
population differences or mimicry. They learn by sorting a large number of
samples into the most likely bins. I recently tested a device at my home,
called BirdWeather, that does similar things; at first it was wildly
inaccurate, giving me species that would trigger a lot of excitement if
they were found in this area, but over the course of a week it improved,
zeroing in on the common species in my yard. However it persistently told
me there were Cedar Waxwings present, which do come through once in a while
but were definitely not present that week. It never detected Bewick's Wren
although they are always here; possibly because of the complexity and
individual variation of their song? Don't know about Merlin per se but it
would have the same issues although possibly a better performance.
Bottom line; you need to verify these audio ids, particularly in a new
location or in the case of migrants moving through.
Jan Leonard
On Wed, Mar 25, 2026 at 9:35 AM Naomi Goodman via groups.io <nlgoodman=
<hotmail.com...> wrote:
> I had the same experience at Burleigh-Murray earlier in March. I've
> attached a recording of a hummer that Merlin IDd as Rufous. It had a mostly
> rufous back with a small green patch. It doesn't surprise me that Merlin
> can't distinguish these two species calls. The sonograms of the
> "Dzeet-Chippety" calls in Peterson's Guide look very similar. If the calls
> are variable, it might be a tall order to improve the ID. What
> characteristics do you use to distinguish between the calls?
>
>
>
