Pinned post
Just a tiny fraction of my favourite moments from 20 years of presenting @BBCClick. It's been an absolute blast. Thanks for watching, and I will definitely see you soon!
5:24
@spenley
Pinned post
Please allow me to bring you some Christmas joy - it feels like we need it...View quoted post on X
Since I said publicly that this year’s new Christmas songs all sound like the lyrics are straight out of a poor AI, I thought I should see if I could do better. And even if I do say so myself, I made an absolute banger!
I asked Gemini to create the lyrics for CLAWS’s Christmas hit, and it came up with something that I think is far cleverer than some of the human-written cliches out there.
I love the bad-girl rock vibe in the lyrics - Lydia spiking the eggnog, getting up to no good, Westa rocking the neighbourhood. And then a clatter on the rooftop - “not the gentle pitter-patter of reindeer touching down”, no, this is a rock and roll sleigh with flames painted on the side!
I fed the lyrics into Suno, asked for Christmas instruments and a Christmas chord structure, and I have to say it came up with a melody that flattens everything else within a 50 metre radius.
It has energy, fun, all the Christmas feelz, and loads of the little fills, ad-libs and riffs - Christmas bell solos, and that little break at 2’18, when the instruments drop. Brilliant!
But all of this brings me to the huge technical nightmare of finishing this song…
GEEK DETAILS
This is not the original version of the lyrics. The song came out pretty perfectly formed, but the original chorus contained two lines I wasn’t happy with.
The original first line was “forget the tinsel, forget the bows” which looked fine written down, but Suno pronounced “bows” as “boughs”. This sounded stupid and didn’t rhyme with the next line in any way.
And the last line of the chorus was originally “Oh, Santa CLAWS is coming tonight”. It was only later that I decided I wanted to get clever and play on the double-meaning of the band’s name. “Oh, Santa, CLAWS are hitting the town” was far cooler.
But herein lies the problem with generative AI - making slight changes to an otherwise great creation is almost impossible, as it seems to start again and make something new. Hence when I put the new tweaked lyrics back into Suno, I got a completely different tune.
Suno does have a song editor however, which, in theory, allows you to rewrite or change small sections of a song, white keeping the rest of it the same.
In theory.
I successfully changed the word “bows” to “stars”, which was simple enough for it to re-sing without messing anything else up, but could I get it to re-sing “is coming to town” as “are hitting the town”?
No.
Nope.
And no again.
With every failed attempt it proceeded to mangle more and more of the surrounding song, and I was left with several Frankenstein’s Monsters of garbled half-lines, non-words and random fills, oohs and yeahs.
I eventually managed to get it to sing the new line three different times, across multiple versions, none of which could be used in their entirety.
Fortunately Suno has one other feature - it can give you stems. These are the individual audio tracks - lyrics, instrumental, and it can even break out the solo instruments if you need them. Weirdly though, it doesn’t seem to know what these are until you ask it for the stems, at which point it seems to listen to the track, and isolate them using audio trickery. This means the audio quality is not perfect, with each stem having artefacts, glitches and echoes of other tracks that Suno hasn’t been able to completely get rid of.
My suspicion is that Suno doesn’t compose and create a song track by track, as you might in a studio. It really does plop it out fully-formed and pre-mixed (which is actually even more incredible if you ask me), and if you want isolated tracks, it has to work them out afterwards.
So this song, “Santa, CLAWS Are Hitting The Town!” is actually my own audio mix of the instrumental track, most of the original lyric stems, and the resung lines dropped in at the right points.
I think the audio quality has suffered, but the overall vibe is still great. The lesson - be absolutely sure of your lyrics before you press the Create button!
AI’s $3 trillion betView quoted post on X
@spenley joins @suziperry and @JasonBradbury on the podcast this week and explains how AI infrastructure commitments since 2013 have been estimated at around $3 trillion — with another $3 trillion potentially coming by 2030.
Revolution or bubble?
Watch or listen to the full episode now:
linktr.ee/thegadgetshow