Plain digital audio files, such as recordings of a voice, are like cassette tapes — a recording of how something sounded. Digital MIDI data, on the other hand, is more like sheet music: a set of instructions. Its power lies in a producer’s ability to assign notes to a new instrument, change their pitch with a click, and other fanciful stuff.
Several iPhone apps bridge the gap from audio to MIDI, doing their best to convert a signal — like me humming into my phone — into a MIDI part.
Why is that cool? Because if it worked perfectly, with a little practice, you’d be able to write parts for any instrument by simply importing the resulting MIDI file into GarageBand or anything else, and assigning it to any instrument. With little other musical know-how, you could theoretically lay down a saxophone part, or even an entire symphony of sound while waiting for a meeting.
Granted, perfect audio-to-MIDI conversion remains a huge challenge for pro studio software, so we didn’t expect much from the following audio-to-MIDI iPhone apps. The Sonuus B2M hardware converter costs $100, by comparison, and can only understand the relatively clear tones of a bass guitar, and not the intricacies (and imperfections) of a human voice.
So what hope is there for an app for the rest of us, that costs just a few bucks and runs on the iPhone, iPod Touch, or iPad, to transcribe accurate instrument parts from the human voice? I found out, by testing the only three iPhone apps around that attempt to pull off this near impossible task.
Results ranged from unusable to the piteous, although the apps perform markably better in the video demonstrations with instruments (scroll down: 1, 2, 3) than they did in our human voice test — which admittedly didn’t have gaps. We’ve included audio samples of our original humming below, as well as the results of our experiments, for your elucidation and amusement.
SongCatcher comment: “Please watch this to see Audio to MIDI conversion on SongCatcher. [It] clearly describes and shows how to use it, and if you use it like this, it works. The audio file in your review above has no gaps between notes, making it hard for the converter to discern any notes, hence there is no output. I’m not saying it’s perfect, but it does take you straight to the midi editor for corrections after your performance. A little bit more effort and maybe you wouldn’t slam a product that you do not know how to use.”
The short story (updated November 20): For now, the voice contains too much variety and dissonance for today’s mobile technology, so the idea of recording complex MIDI parts from one’s humming is still a bit out of reach. Basically, we’re still waiting for one that’s specifically designed to deal with voice rather than instruments. For those who play instruments, these apps are a neat way to lay down music for any of virtual instrument, using the real one in your hands.
Here’s the original melody I hummed. It contains some quicker and some slower sections:My Humming
Most Accurate: Magic Stave
By far the best of the three apps we tested, Farnell Computer Services’s Magic Stave ($3, pictured below) has a trick up its sleeve that we suspect is the “key” to its success. It guesses the musical key of the hummed tune, then detects only notes in that key. This is a really clever approach, because it filters out extraneous notes (unless you’re working a weird key signature).
Here’s what its version of our humming sounds like with the Magic Stave-produced MIDI assigned to a piano sound:Magic Stave's Version
The result bears some resemblance to the original melody, which is pretty impressive.
Then again, if you wanted to use this melody in a music project down the line, you’d have to spent so much time tidying up the MIDI that you may as well have re-played it in on a keyboard — or even drawn the part into a sequencer.
But it’s close-ish — and, perhaps more importantly, it’s much closer than the rest.
It was all downhill from there, for the most part. Songcatcher ($15 or Lite/free) picked up almost nothing of my original hummed melody. Intelligent Gadget’s Audio to MIDI Recorder ($3, formerly MIDIRecorder) had the opposite problem, picking up so much background noise, even in a quiet room at home, that the result was garbled beyond recognition:Songcatcher's Version
In its defense, Songcatcher (pictured left) has a nice-if-small sequencer and audio mixer, and if you’re trying to lay down a melody and zap the audio to your computer(s) easily, it does that just fine. But the hum-to-MIDI feature we are interested in for the purpose of this article was essentially useless.
Meanwhile, Audio to MIDI Recorder (since updated) was the buggiest and least stable of the lot.
Whether or not you call these audio-to-MIDI attempts valiant or futile, spending too much money on them as things stand now in the hope of churning out a sparkly, perfect MIDI file you can use without further editing, is just too much to ask. Unless you’re willing to work with results that add input from a “ghost in the machine,” save your money — and console yourself that there are still some things the human ear and brain can do better than a computer.