Voice Over Talent

Posted: November 29, 2023

SAG-AFTRA in 2023: To Ratify or Not to Ratify

I believe in the potential and power of advancements in computer science to help us solve big problems, especially in industries like healthcare, energy, and manufacturing. I believe in mobilizing around problems that are almost universally recognized. I’ve invested in several AI funds and companies, ranging from those focused on software (APIs, data sets, and language models) to hardware (processors and ways of building them), for this reason.

I’m not a big believer in creating solutions and then trying to find problems to match them to.

I get the argument of “If they use your digital likeness in ways to which you’ve not consented, and we catch them, we’ll sue them,” but it has two vulnerabilities. First, for rank-and-file, i.e., non-celebrity, actors who already sometimes aren’t given both copy and credit for their work, and can’t find it when they look for it publicly, that can be a big “if.” Second, it doesn’t account for amalgamation. That is, generative AI can be used like a blender. Imagine…

FacialScan && VoiceMatch(MEDIAN(“HumanActor1”, “HumanActor2”)) = VirtualActor1;

So, a hybrid, digital-only performer, results. Where will metadata ensure that both HumanActor1 and HumanActor2 will always get paid, commensurate with whatever use of VirtualActor1? Also, how will such usage be made to always comply with the rights granted by HumanActor1 and HumanActor2 at an actual technology level? Where is the enforcement apparatus that ensures VirtualActor1 will become a recurring revenue source for everyone without being applied in a way that could negatively impact the human performers’ brands or reputations?

There is no such system. That boils it all down to a matter of trust. When the technology can be used in so many “Who’s gonna know?” ways, that’s a pretty tall order.

In the meantime, actors are now getting asked to audition, and then, if booked, we’re to get only 48 hours’ notice before we work, within which to give informed consent to the application of AI to our performances or vice-versa. That pits us against one another over the issue. It dangles work in front of the performer and then pressures the performer to consent. If they don’t, they risk losing the work to another performer via re-casting at the 11th hour. The performer who then gets released collects no cancellation fee. All of it is a far cry from what should be the protocol:

Per project, disclose intent to use AI within the auditioning process so that actors who don’t want it applied to their performances can choose not to waste their time and decline to audition in the first place!

When DRM got hacked out of the original MP3 format by an anonymous “Information wants to be free” punk in Australia, which wasn’t responded to swiftly and decisively by the global community (Don’t get me wrong: Personally, every Australian I’ve ever met, I’ve always adored and still do), where did that take the music industry? By the time streaming became a thing, the market value of music had gotten driven so far down that the streaming services just capitulated. The argument being, “We can’t stop piracy. At this point, it’s endemic, so we have to treat the value of music as… ‘ephemeral,’ not totally worthless, but more subjective than ever.” They priced themselves so low that the compensations for musicians became so, so much worse than they’d been when the delivery mechanisms were physical media (CDs, CSs, vinyl). As artists saw their livelihoods suffer or get extinguished, to profit as much as possible as the new corporate intermediaries between creators and consumers, they did what they felt they had to.

So now, generative AI is here. Some applications of it can be laudable and miraculous. Others can threaten not just content creators’ livelihoods, but national security. With technologies that behaviorally still aren’t yet fully understood, and aren’t yet fully under control, and can output things that can’t be fully tracked and audited on-demand, let’s not get ahead of ourselves. Let’s avoid making the same mistake again, this time a thousand-fold.

Anyway, back to the matter at hand, strictly from an actor’s perspective:

On one hand, Val Kilmer gets his voice back. On the other, generations of performers are now being positioned to choose between landing a given gig and having sustainable careers. In areas otherwise such as residuals, pensions & health, performance capture, and equity & inclusion, real gains have been made, by which I’m impressed and for which I am thankful. Again, props to the negotiating committee on wins realized on all those other fronts. There has been good progress made regarding AI considerations, just not enough. On ratifying the overall new tentative TV/Theatrical agreement, therefore, I’ve voted No.

Resources

Posted: April 24, 2023

The 2023 WGA Strike: A Lesson in Trade Unions vs. Other Types

A good colleague once confided in me at one former employer that he felt unions were obsolete. He believed that while they helped build this nation’s Middle Class, they were no longer helpful.

I’ve always disagreed. Unions are like for-profit and non-profit corporations and just about any other institution. They can be run efficiently, in competitive ways, and in line with industry trends and needs, or they can be out of touch, bloated, and corrupt.

While the pending WGA strike, which SAG-AFTRA now supports, will continue reverberating throughout the media and entertainment industry, I’m among those who don’t expect it to last long.

The gradual recovery of unions here in the USA, at least in some industries, is a good thing that gets a good amount of press.

However, the difference between labor organizations at the company rather than at the sector level needs to get more attention. America needs to learn more from Europe in this regard. “Labor Union” means organized workers, whereas “Trade Union” is more specific. Using the terms interchangeably can be a trap.

When employees of a given company organize, they need to negotiate just with the management of a single, specific enterprise. When workers collect across a whole sector, they deal broadly with the management of companies that compete with one another.

That’s a huge difference. It can make talks between management and labor way more productive. It helps maximize partnership and keep the relationship in between from becoming adversarial. By design, it makes strikes much less likely to happen in the first place and keeps them from drawing out if and when they ever do. When a whole industry gets locked up, there’s much more urgency to find common ground so everyone can return to work and business can resume normal operations.

Structural details matter. I’ve yet to operate out of Europe or any other continent. As an American laborer, I’ve been proud to be a member of a collective bargaining organization as a freelancer (actor). Whenever working as an employee (tech worker), I’ve never been a union member. In my world, the two concepts have never conflicted or been in contradiction with one another.

Apples and oranges. Sure, work is work, so ultimately, it’s all fruit. But the details make the difference in whether everyone has enough to eat.

Resources

Posted: April 10, 2023

AI in VO: How to Get it Right

A couple of friends shared spot-on points in this article, and someone asked for my take on it in a recent meeting. So here’s my 3rd cent on the matter.

I don’t think I’ve ever seen this topic become as active as I have so far this year, and I have a confession to make:

Some years ago, I got booked on a “research” project for one of the companies developing AI for use in voice applications and otherwise. Was I well paid for my time and labor? Yes. Did I work with great people and have fun doing it? Absolutely. Do I regret the job? Yep, because, in hindsight, I went into it signing away rights to more than I should’ve.

At the time, I was still a Non-Union actor. I was also a full-time freelancer, as I am today. Halfway through the project, I started thinking, “Hm, am I putting myself out of a job here?” As many people do, I fell into the trap of taking short-term gains without fully considering the potential long-term consequences.

In the VO business, issues started with high-speed Internet connectivity, creating a competitively over-saturated buyers’ market. Now with AI, humans are being removed from the equation: Innovation brought more people into the business than it would ever have enough work for, and now it’s started to take the work back and away from people altogether. The Tech giveth and The Tech taketh away.

Shiny new toys can have a way of turning businesspeople into overthinkers and also a way of turning artists into nerds. For Producers, the trap of AI voices, whether “cloned” or wholly synthesized, is over-engineering and going down a road that may be penny-wise but pound-foolish, especially if using entirely fake voices. There’s no point in spending time and effort on creating something sterile that doesn’t make the audience feel something, engage and take action.

It doesn’t have to come to that, though, nor should it. Hollywood and Silicon Valley must keep learning to speak the same language to avoid this and find a sustainable middle ground.

For performers, as to AI as to lowball rates. Professionalism means knowing how to do business, think creatively, and negotiate, but also when and how to refuse to even audition. Always be willing and able to say no; if needed, walk away and do something else for a living.

If a company wants to “clone” your voice, then every time that clone gets used in something, how and for how long, you need to collect commensurate fees. This business has standards for a reason, and applying them in a way that keeps pace with innovation is just an Engineering matter. AI companies doing business in good faith should relish the opportunity to do so if they want to gain legitimacy and take market share.

Example 1:

Advertising; a wild spot, running on Chicago radio for (1) cycle, i.e. (13) weeks

For that, Scale is currently USD 429.19. So when creating the project session on the AI platform, the buyer needs to specify that as the intended usage and then pay that, plus additional one-time fees of 17.25% for health insurance and retirement contributions and 6.75% for employer contributions. Once they approve that cost via the AI platform, then the buyer can create the spot.

What if they want to be able to make multiple cuts or edits afterward? No problem. If they want to do so before the cycle’s End Date has passed, an additional $429.19 must get paid to the talent every time they export (or “bounce”) audio from their account on the AI platform. If they want to go back and make edits after the cycle’s End Date has passed, fine. That should be the same process through which they could optionally extend or change the usage. In that case, just like before, the AI platform should recalculate the fee, get their approval to let them back into the project session, and then charge them that fee when they save their changes and export the audio.

Example 2:

E-learning

That’s non-broadcast corporate/industrial work, Category 1, i.e., “Programs designed to train, inform, promote a product or perform a public relations function (which) may be exhibited in classrooms, museums, libraries or other similar locations. Included are closed-circuit television transmissing and teleconferences.”

Up to an hour of work would be $559.24 between the session and H&R contributions, plus a few percentage points for one’s paymaster. So, to use a “clone” of your voice, the buyer should pay that as the initial base fee, and then any time they want to go back into the AI platform and make changes, additional costs need to be paid to you based on the exact word counts of those changes.

This versioning and management functionality has long existed in CMS platforms and applications like Adobe Acrobat. Comparing whatever edits Producers made to an original script is a “DIFF” operation software-wise. It doesn’t require anything so advanced as machine learning or AI, so there’s no reason any company selling AI voices shouldn’t be able to implement it. $0.20 – $0.35 per word is standard. Buyers should still pay for every change, but only for what degree of changes get made each time.

The Bottom Line

AI isn’t going away, and for buyers who want to settle for it, it can give them tons of ease, flexibility, and convenience. It can also help them save money in the long run. On the talent end, it can save time and labor without talent getting cheated out of the recurring revenue it still deserves. Like anything, there are trade-offs all around, so everyone should keep an open mind.

As we actors go, how many of us behind the mics wouldn’t prefer to focus our manual energies on auditioning for or working on commercials, promos, and character work, where it’s all about the exacting creative details and individual performers’ ability to take direction, improv, and act all in real-time? As actors, aren’t we supposed to be storytellers who make audiences feel something? When clients on some other job types often want something low-cost, fast, and with utility, “a nice sounding voice” as a relative commodity, why shouldn’t we be optimizing while still monetizing? Win-win here is possible if both the Buy and Sell sides genuinely want to dance. It takes two to Tango. Anything less than that becomes just more “industry” pain and balkanization.

If you’re genuinely professional talent, you owe it to yourself and your trade brothers and sisters today and tomorrow to be innovative. Make hard choices as you have to. It’s vital to consider when and how to turn down a job to protect your career. Live high or live low, but never compromise your principles because your choices will always stay with you no matter how you live or where you go.