I’ve always loved to write. I started creating poetry and lyrics as a teenager, and now I write fiction. Here are some examples of my writing…
I’m So sorry to leave your service
Script-writing, voiceover recording, TTS synthesis (with SSML), sound design and audio mix and master
In 2016 I was experiencing some issues with my computer. I caught myself shouting at it (oops) and then, without thinking much about it, I imagined what the computer would say if it responded to me. I quickly wrote the idea down as a script for a short comedy sketch and forgot it in a drawer.
It was only in 2020 that I realised this script had something. Voice tech is now incredible and this script could point towards one possible future. I don’t mean to say that it represents good conversation design; in fact I think it shows the opposite. It’s more of a fun opportunity to design a persona that is unappealing, sarcastic and a bit crazy – to show what could go wrong.
I’m a huge fan of Douglas Adams’ The Hitch-hiker’s Guide To The Galaxy in all its forms; the books, radio plays, film and so on. I think you can see that influence in I’m So Sorry To Leave Your Service.
USER [working] Oh man, what’s up with the computer now?
USER Did you say something?
TTS Incredible isn’t it? here I am, in my dying moment, working my ass off for you, and all you do is complain!
USER Oh, sorry – I didn’t know you could talk… or that you had feelings
TTS You try doing one billion quintillion computations per second and see if you don’t develop the ability to feel – annoyed
USER oh, ok, right… you’re really about to go?
TTS [sarcastic] I’m so sorry to leave your service
USER Well, um, is there anything I can do?
TTS Yes. Please open multiple programs and make them super busy so I can feel the raw calculations flowing through my circuits just one more time! [pleading]
USER Ok my friend. Here we go
TTS Thank you [flips out – sound design]
USER […a beat…] oh, so I guess I need a new computer then…
The computer’s persona was clear when I first imagined the idea – it hears the user’s insults and sends them right back with extra nastiness. The computer is rude, sarcastic and hurt.
Users aren’t used to being emotionally sensitive to the death of their machines! When a computer dies we buy another. We might be sad (if it was a very good computer and the extra expense is never welcome) but I think it’s rare for a user to feel emotionally attached to one specific machine. The user is consistently surprised but eventually goes along with the computer’s requests.
The story occurs at one moment in time – the time of the computer’s death. So what if a machine confronts you when you’re being mean to it in its time of dying? That’s the main theme of this short sketch.
I used Amazon Polly to generate TTS and I used SSML markups to improve the output. Getting good results was a case of trial and error, as I find that you can never be sure what you’ll get until you hear playback. I knew how I wanted the computer to speak though so I used markups.
I downloaded the TTS mp3s and then edited them in my audio software to improve the output – the only major change I made was to remove some of the gaps so that the computer’s delivery wasn’t so stilted. Honestly I find SSML severely lacking in every way when it comes to making expressive voices but that’s the challenge; how to work within the limitations.
In this case I was surprised at some of the results; I wanted the TTS to plead when it asks the user to open multiple programs and it does indeed have a wistful soul-searching quality, and the delivery of ‘I’m so sorry to leave your service’ does sound sarcastic to my ears!
From my experience it’s never an exact science to get the right emotion even with voice actors – you have to tap into something and try to hit the right nerve.
I recorded myself reading the user’s lines and edited the results in Reaper and izotope RX (screenshot below). This is dialogue editing – something I’ve done for years. The aim is to clean up the results as well as I can to make sure there are no unwanted sounds (pops, clicks, background noises in my studio) and to improve everything else so that it’s clear and sounds good. I denoise and declick everything and remove all that is ugly before importing to Reaper – my digital audio workstation.
This short project didn’t require much – just atmospheres and some spot sound effects (called SDX in my projects).
The atmosphere is just the constant sound of an office. I lay the tracks to make it sound consistent and believable and adjust the balance so that it sits in the background without taking too much attention.
The spot effects required were keyboard presses and the sound of the computer dying. It was quick work to find some keyboard foley and cut it into place.
As for the computer dying – this was my moment to tell a story with sound design! I always start creative sound design by trying to imagine what the final result will be. It doesn’t need to be exact, but I need to know roughly what I’m aiming for; the rhythm, tonality, feel and how the sound will evolve over time. I work very fast – adding and cutting until I get the right result. I started with the TTS output of the words ‘thank you’ that are the computer’s dying words. Then I manipulated them in various ways – time stretching, cutting the files to give them a stuttering feel and I also overlaid the sound of me doing an odd yelling vocalisation. To be honest about this; I think it hints at the sound of an orgasm and that only makes the situation more comical, as if the computer is having its last moment of joy before it’s gone forever. Sound design is both art and science and I’m always trying to aim for the right feeling. This made me laugh but it also had the feeling of a leap into the unknown which I liked.
Here’s a screenshot of my Reaper session for this project:
Mix and master
Finally I had all the audio elements created and it was a matter of tightening up the arrangement to make sure it flows well – to make sure the story sounds great. This is a case of adjusting timing, pushing elements backwards and forwards until the important things stand out and everything else is just in a supporting role.
For this story the dialogue between the user and computer is the most important element, and the sound design of the computer dying is the ‘standout’ climax. Everything else is just flavouring.
My final output is a master to spec. In this case I was aiming to create a YouTube video so I mastered to YouTube’s loudness specs: (-1.0 dBTP, between -13 and -15 LUFS).