What is Microsoft’s Songsmith like? Enterprise software, that’s what

There’s a powerful meme going around, especially on Twitter, pointing to examples where people have taken the vocal tracks of famous songs and then got Songsmith to write the music, which is what it’s designed to do.

So you get Stairway to Heaven, Roxanne, and a whole stack here by Dan.

Lord, it’s horrible. Which led Justin Williams to ask

WTF was Microsoft thinking with Songsmith? Here’s it does White Wedding by Billy Idol. Just unbelievable

Yes, what was Microsoft thinking? Well, let’s start. Someone thought “Apple does a music app.” (Garageband, below.) “We should do a music app. Apple’s one lets you create stuff. We should make it easier. We should write the backing – we’re smart. OK. Tunes follow a structure.” And then “what people sing follows a vague structure.” And then “we can fit the tune around the singing.” And then “what they sing becomes the structure.”

It’s enormously clever; but as the examples all show, utterly stupid. Songsmith has no notion of what a great melody, nor a great accompaniment is; indeed, it doesn’t understand melody, only the idea of progression through a structure. What the person is doing becomes essentially irrelevant; they’re just an input. Listen to enough of these ..creations and you start to notice a certain sameness to them that isn’t there in the originals (obviously). Everything is hammered flat. The surprising harmonies of Sgt Pepper’s Lonely Hearts Club Band? Not got those. The plangent arpeggios of Stairway to Heaven? Nope. The buzzsaw guitar of Jonny Greenwood, determined to break up Thom Yorke’s sweet-sounding chorus to Creep? No idea what you mean.

And to answer Williams’s question, what was Microsoft thinking? It was thinking what it always thinks. Reduce the human element to an input, put it in a box and make everything exist only in that box. Remove the space for human creativity that hasn’t been thought of already by the programmers. Think inside the box.

It’s pure enterprise (as in, big company) thinking, applied to one of the art forms that has been with humans for millennia. No wonder people are astonished and can’t stop pinging it around the intarwebs (for it has to be said, Songsmith is getting the most fantastic publicity – you’d think it was an Apple product).

It’s intriguing. Apple has Garageband, which is a tabula rasa, the original blank slate, that offers you fills and guitar and piano twiddles, but you have to do the creative act of putting the song together. (I’ve always thought it makes it too hard to create a long piece; why can’t I just say “I’d like to have something four minutes and 30 seconds long with this drum track, set it up please”?). Even pros use Garageband.

Songsmith takes away the blank slate. In its place it… tells you what it thinks you’re thinking. It is scary. I’d love to know what Microsoft’s engineers really thought before they released it into the wild.

And, of course, whether anyone will release a song “written” by Songsmith. Something tells me not.

6 Comments

  1. Band in a Box has a feature that lets you generate a certain amount of music based on a style. It’s pretty good. (But there again BIAB generates solos that are better than many that get recorded by real musicians!)

  2. Great post, Charles. What the hell kind of review or QA or beta testing does MS do? None, I’m forced to suggest because if they did any at all, this appallingly misconceived idea would never have got out the door.

  3. According to Leo Laporte, the commercialisation and video ad for Songsmith was made paid for privately by developers of Microsoft Research Labs. Microsoft didn’t want to bring it out.
    It has gone viral, which is good for the product. I agree it is horrible.

  4. Thanks for linking to my videos!

  5. You’re being too harsh. Listening to the Band-in-a-Box Eurodisco rendition of Wonderwall, it’s clear that this software is going to save Eric Prydz a shedload of time. Right now, producers have to time-stretch and beat-match the vocal sample to the backing track: this lets you do the opposite. So expect many more four-on-floor renditions of 70s classics.

    The problem that Songsmith has is that it’s primarily a research tool. It exists to show off a technology that would otherwise would wind up absorbed into something like Cubase or Ableton Live. It’s not far removed from what the Yamaha PSR keyboards or Bontempi organs of old except that it has a lower bar of entry – removing the need to know which chords you want.

    This is what the researchers wrote last year: “No one will ever write, record, and produce a top-40 hit with any version of MySong. What MySong does do is give many folks who would never even taste songwriting a great opportunity to just get a glimpse of music creation.” [It got renamed to Songsmith on the way]

    The software does a pretty good job of working out which chords fit a melody. I’d argue that someone doing a first cut at harmonising a melody would probably come up with something not far from what the tool does – uninspired choices but choices that fit a known structure. But that’s where the difference with human composers come in. No-one ever sends off a song with that kind of harmonisation. It’ll get worked on a lot and, with any luck, come up with something that isn’t your regular ii-V-I style progression. You can play with the progressions in the software to give the progression more of a minor character (the unhappy-happy slider) or use extended chords (the jazzy slider) but right now that’s all you can do. Software like Harmony Navigator or BiaB is going to be a lot more useful to someone wanting to do more than a quick reggae cut of Happy Birthday. I guess that’s why PG Music got involved in the project.

    I think this exercise will probably rebound worse on PG than on Microsoft. At least Songsmith can be parcelled up into a teaching tool once the meme goes stale. People are going to remember the the ‘cha-cha-cha’ backing from these videos and think “BiaB – oh yeah, that’s the Bontempi backing software”.

  6. I should note that there’s a considerable difference between the Classic Hits by Microsoft Songsmith I’m producing and the others on YouTube — I postprocess the MIDI file IN LOGIC, because Songsmith not only can’t match tempos (and doesn’t provide for decimal bpm values) but can’t even keep its own tempo. The tempo map for a typical song I make has at least 8-10 tempo changes throughout, though this is a combination of the factors of inconsistent drummers (even the great Neil Peart wasn’t rock-solid on Tom Sawyer) with the inconsistent/inaccurate tempo provided by Songsmith.

    What’s more interesting to me is how much criticism the instruments get for sounding so fake and MIDI-y. Yet I don’t use the Songsmith instruments, I use professional (since I’m an actual musician) software instruments in Logic, instruments that when provided proper MIDI values sound nearly indistinguishable from live ones. Songsmith manages to output such robotic MIDI that it dehumanizes (not the right word, but I’m looking for the reverse of ‘humanizes’) the natural-sounding instrument.

Comments are closed.