Thought about this for a while and finally came down on this division: You gotta start by breaking the entire set of sounds into groups. Usually called "range" When considering the frequency range we have to agree on some terms..but you probably have already heard them. Bass sounds are the lowest, Tenor is mid low, Altos are mid high and Sopranos are the highest. The way to think of this visually is by looking at a piano. Humans generally can hear in a frequency range between 20hz (left end of the piano) and 20,000hz (or 20kHz, the right end of piano) with a 440h "A" note near the middle of the piano.
This classification increases along an exponential curve, which is why a piano with 88 keys represents a difference of around 19,000hz, rather than about 100hz.
You can look at a visual EQ monitor in iTunes..or on your dad's 70's era "receiver" that he still keeps plugged into a 5 disc changer, despite the fact you keep buying him the newest iPhone.. and get a quick understanding of how this works. If you turn the two left-most EQ knobs all the way up and the rest all the way down, you'll hear a lot of Bass and probably mess up your speakers pretty quickly, but you'll lose practically everything else.
... and at the risk of stating the obvious, we know the low end of the frequency range is bass sounds, but in an orchestra that might include Tuba, Bass Clarinet, Baritone saxophone and the kettle drums. The very high end frequencies will be flutes and clarinets, trumpets or even higher pitched brass horns, cymbals etc..
A dog whistle is pitched above 25khz and usually closer to 40kHz...which is why it will never bother you but it will send Spot scrambling in circles. If you DO find you can hear a dog whistle, check with your doctor to see if you're turning into a Werewolf.
This division along frequency range might sound a bit arbitrary, or at least not specific enough for learning to distinguish instruments, but you'll find that training your ear for timbre (an instrument's specific quality) is a lot harder than training your ear to notice frequency. Once you learn to recognize frequency range (and you'd be surprised how with a little bit of training you'll quickly be able to pick out notes AND their exact Hz) you'll find you can easily separate groups of instruments IN TO range .. and after THAT you can start distinguishing the instruments themselves.
Personal Anecdote feel free to skip
I started training as a violin player at age 3, and I was taught by the Suzuki method which wasn't that common when I was taught. Suzuki method worked thusly: The teacher would play a phrase. The student endeavors to play it back with no other visual cues (written music, etc) The crazy thing is I hate the violin now, but I've been trained on several instruments and that ability to repeat a phrase after hearing it once has never failed to make learning a new instrument easier. I can't really know if other ways are better, but I think it helped my music career.
Years later, studying audio production, the only ear training method that all recording students were graded on (regardless of whether you played an instrument) was called "GoldenEars" and it was based solely on frequency recognition.
So that's my story, and I'm stickin' to it.
PS. =Can Software Pull the Saxophone out of an Ensemble Recording?=
I really like the concept of software that could pick out an instrument from a recording, and I imagine it isn't as far off as we might think, the codecs that allow us to make mp3 files out of much larger "real sound" WAV files could be viewed as the opposite of this, as they work by removing bits of "redundant" information from the larger files. Most people now 20 years on from Napster can hear the difference between an MP3 and a WAV but there's an accepted level of decay that most people are willing to live with. If we can program the removal of a certain percentage of information in a sound file and still hear basically what we began with, I have to imagine that pulling a certain classification of samples and bits could give you just vocals or just instrumentals. It would also depend a great deal on the type of ensemble. Pulling just JOhn Coltrane out of the Miles Davis Quartet wouldn't be all that hard, as the Tenor sax is easily distinguished from Miles' Trumpet and Paul Chambers drums. But an orchestra situation obviously creates new hurdles. Regardless, the people who will crack this will of course be DJs. Necessity is the mother of invention and DJ's are constantly looking for instrumentals and Vox-only tracks... So, as usual, the street will break it, probably via something we already have access to, and then ProTools will quickly commodify it and we'll pay big bucks for said filters. You heard it here first!