Video producer Jo Brind, creator of the AI -performed realisation of the play Fly Me To The Loon, shares further details of how he managed to transform the script into a performance.

To begin with you need to start by getting an AI human voice generator so that you can create your performers. You can find these on google and there are several to choose from.

The best tip I can give you is to do a lot of work on the script right at the start. Have a clear vision about the play and decide how you want the characters to come across. Get it right and it will pay you big dividends. I say this because I came to the conclusion, rather late in the process, that one of the voices was too soft. I felt it needed to be tweaked just a little.

If I had realised this at the start, I could have changed the whole voice output with a couple of flicks of the controls. As it was it was, it was a case of going through the whole thing and changing each bit individually.

And don’t get too downhearted. There will always be unexpected problems. For example, working on the script by Frank Gauntlett -the noted Australian playwright whohappens to be a good friend), Frank uses conversational idiom. Instead of saying “I live down the road” his characters will sometimes say “Live down the road”. Unfortunately, the program interpreted this as LIVE (as in live music or live performance) down the road. That was quite difficult to deal with.

And Frank’s play is about the MI5/KGB double agent Michael Bettaney, who in his youth had a habit of goose-stepping round the pubs of Oxford. At the time he thought he was a fascist. During the play he and other characters mention the Nazis. This was a problem for the program. Some voices could say Nazi, others couldn’t.

You will also need to go through the script and break down proper nouns into recognisable parts. For example the script referenced the Bodleian Library. The program didn’t seem to know anything about this, betraying the probable American ancestry of the software, and so I had to assemble it from word parts (Bod-lee-Anne). And generally, I thought it was poor at English place names.

The program offers some refinement in the way it renders speech. For example, you can highlight words and then have them rendered faster than the remaining text, but generally, I did not feel comfortable doing this. It didn’t seem to make much difference, and I thought I could do a better job in my video editing suite. For example, I got one of the characters to sound progressively, increasingly, drunk by slowing down his speech and making the voice more base. This is fairly easy to do in Final Cut Pro. But it was laborious, because I had to make slow changes to the voice, but I don’t think there is a better way to do it.

I also had to think about background sound. The play is set in a pub, so it seemed appropriate to have the murmur of conversation behind the characters lines.  I recorded some live voices at a party and used this hubbub as the background. I thought this made the play more lifelike than Siri or similar characters, but people I spoke to said they still found the voices unrealistic. This may, however, be because they knew it was AI generated before they started watching.

This is just the first iteration of Fly Me To The Loon. AI is rapidly progressing with updates becoming available all the time and I will keep refining it and I know it will become even more lifelike and realistic. My advice to anyone wanting to have a go at creating an AI performance is to say, do it, with a bit of patience, and some trial and error, you should be able to succeed.

Watch Fly Me To The Loon

Jo can be contacted on jonathan@brind.co.uk

To find out more read the interviews with Jo and Frank here.