New AI tool auditions for gigs using actors’ own voices

The big existential fear on everybody’s mind used to be losing our jobs to AI. That all changed when the pandemic hit and millions were faced with job loss that had nothing to do with automation.

For many, the best solution to their COVID-related employment woes was to turn to the freelance or “gig-based” economy. This led to a huge uptick in value for companies specializing in connecting clients with freelancers, and Fiverr, an Israeli-based company, was no exception.

But the tides are changing, and workers are returning to the office. This has caused a bit of cautious pessimism for the market and even Fiverr, whose stock peaked just last year, isn’t immune to the effects.

In order to get things moving in the right direction, the company’s rolling out innovative new solutions that it believes will keep workers and their customers on the platform in the post-COVID era.

Perhaps most interesting among these new solutions is Fiverr’s “AI Auditions,” a machine learning-powered tool that synthesizes human speech so that actors can audition for voice roles asynchronously.

Up front: Fiverr’s new tool is a director’s dream come true (if it works). It’s only being offered for use to a select group of workers, so there isn’t much to go on so far concerning its efficacy.

According to a company press release, the process is simple:

Voiceover artists on Fiverr can create their own AI voices by quickly training the Voice Auditions engine. Customers will then be able to easily tap into Fiverr’s marketplace to “sample” a voiceover artist’s voice without having to contact them first or ask them to audition.

After typing a sentence or two from their script into a text box on the artist’s Fiverr profile, the AI will read the language outloud, in the voiceover artist’s voice, giving them an idea of whether or not that person’s voice will work for their project.

Background: Fiverr asserts this is a big problem that needs to be fixed. According to a testimonial included in the AI Auditions press release, the process of recording and sending voice samples for auditions can be tedious and time-consuming.

The main benefits to AI Auditions would be a one-off training cycle and the ability for directors to hear actors speak any lines.

For actors, the one-off training cycle means they can budget a short amount of time when their voice is in peak performance condition and then let the machine handle the rest of the tedium involved in first round auditions.

And those seeking talent can just generate bespoke audition clips at their leisure, which saves them the time of having to wait for updated audition files.

Quick take: On the one hand, this sounds like a game-changer for people who just want to work and hiring agents trying to find the perfect voice for a time-limited project.

On the other hand, it sounds like a dystopian nightmare.

The problem is that AI Auditions sounds too good to be true. If it’s robust enough to be useful — ie, the audio quality is solid and it truly represents the unique voice of a specific human actor — then it’s probably going to be incredibly tempting for shady agents to abuse.

Looking at this from the point of view of a starving artist or a project coordinator just trying to do their best work, it seems like a godsend.

But Fiverr’s going to have the wall the crap out of this particular garden to stop bad agents from simply typing the whole script into the AI and recording whatever it spits out without ever paying the original human actor for their work.

There’s a decent chance that Fiverr’s AI isn’t capable of outputting Oscar-worthy voice acting performances, but that’s not going to matter for every project out there.

That being said, if Fiverr manages to keep the data under wraps and institutes some sort of watermarking or other measures that prevent intellectual theft… this could level the playing field for voice actors who have more passion for their craft than they do time for auditions.