Notes from the Limited Human Interface

A few people have suggested to me that using AI to help generate music is cheating.

At one time, I would probably have agreed with them. But since I started making music this way, I’m not so sure.

I have huge respect for talented people who can write and play their own music and who are lucky enough to have a great singing voice.

Sadly, I am not one of them.

I’m told I have a good ear for music. I have good timing and can hold my own on bass. I used to play keyboards in a band, provided the parts weren’t too challenging. I can strum a few chords on guitar and I enjoy writing.

More importantly, I love taking a song idea and seeing where it goes. Once I have the lyrics captured, I can usually form a pretty solid melody in my head.

Then we get to the singing.

Oh Lord.

I can frighten children and scare cats when I start belting things out.

Ironically, I have a decent sense of pitch and can hold a note. It’s just that nobody particularly wants me to. My voice is reedy, rather nasal and stubbornly resistant to decades of attempted improvement.

So using AI to help “improve” my voice opens up all sorts of possibilities.

It is also considerably cheaper than the impressive array of electronic wizardry I have invested in over the years in an effort to process my voice into something that won’t cause listeners to cover their ears and run screaming from the room.

Frankly, I see nothing wrong with it.

And then there’s the recording process itself.

Over the years I’ve used Nuendo and various generations of Cubase to record and mix my ideas. I play the things I can play and use virtual instruments to fill in the gaps.

Even when I was recording real bands, I would often enhance the recordings afterwards. Add strings. Put a soft pad in the background. Double a vocal. Add harmonies. Replace or augment sounds that weren’t quite working.

I could also program drums considerably better than I could play them.

Some of the real, living drummers I’ve played with have been awesome musicians.

Others have struggled to count to four and maintained a rather distant relationship with timing and rhythm.

Not ideal qualities in a drummer.

Drum machines, of course, can be “humanised” by introducing tiny imperfections. But when required, they are wonderfully reliable at the apparently controversial business of maintaining a beat.

The problem has always been that it is very easy to spot a bedroom recording.

I enjoy mixing and can tweak things endlessly. Mastering, however, remains something of a black art. In the past I’ve used third-party mastering companies. The results aren’t always perfect, but it does reduce the likelihood of a PC being hurled through a window accompanied by a colourful selection of Anglo-Saxon terminology.

The bigger problem, to my ears, is that home recordings often lack that final level of production quality that I want, and that listeners have quite reasonably come to expect.

AI can help provide that.

Once, that is, you have persuaded it to stop making everything sound as though it has been produced by Keith Urban for a Nashville television special.

Believe me, this appears to be the factory setting for a surprising number of AI music engines.

But with the ability to provide audio references, detailed production instructions and increasingly precise musical direction, things start to become interesting.

I can generate ideas, extract individual stems, load them into Cubase and start working properly. I can edit them, rearrange them, replace sections, add instruments, remove things I don’t like, change harmonies, alter the structure and generally fiddle about until either the song works or everybody else in the house has gone to bed.

In other words, it is still a detailed, painstaking and very human creative process.

If I could afford to hire a group of brilliant session musicians every time I had a song idea, I absolutely would.

Sadly, I cannot afford such extravagance.

So my new bandmates and session players wear digital trousers.

And I’m perfectly happy with that.

Because, as the song says:

We may never be famous,
but we’re still singing our songs just the same.

Which gives me the opportunity for a completely shameless plug:

“Not Famous Yet”, the first single from North of Elsewhere, is released on all streaming platforms on 25 September.

As for me, I’m having an absolute blast.

Of course I miss playing live. I miss the interaction with the amazing musicians I’ve been privileged to play alongside over the years. I miss that moment when a crowd gets behind a band and suddenly everybody in the room is part of the same thing.

AI can’t replace that.

But I’m still making music.

Ideas that previously existed only in my head can now become finished songs. I can experiment, make mistakes, change my mind, tear things apart and put them back together again.

And whether ten thousand people listen to them, ten people listen to them, or nobody listens at all, making them makes me smile.

That’s priceless.

Cheating?

I don’t think so.

And you know what?

I don’t really care.