Episode 20: "Interview with Sarah Fairweather, Senior Program Manager of Ethics at WellSaid Labs"
Listen to this episode:
Episode description:
In this episode I have a conversation with Sarah Fairweather, Senior Program Manager of Ethics at WellSaid Labs, an AI voice generation platform that turns text into voiceovers.
WellSaid Labs: https://wellsaidlabs.com
Episode transcript:
LORENZO: Hello, everyone. Welcome back to AI Quick Bits: Snackable Artificial Intelligence Content for Everyone. Scot Pansing is still the host, but since this episode is an interview with Sarah Fairweather, Senior Program Manager of Ethics at WellSaid Labs, Scot thought it would be fun to have me, Lorenzo, a voice actor on the WellSaid platform, do the introduction. WellSaid Labs is an AI voice generation platform that turns text into voiceovers in seconds. Scot chats with Sarah about her background and how she got into AI ethics, and how the WellSaid platform works to benefit both customers and voice actors. They also discuss safety, deepfakes, and more. Enjoy the episode.
SCOT: Hi Sarah, thank you so much for coming on the podcast. I really appreciate your time.
SARAH: Hi Scot, thank you for having me.
SCOT: Absolutely. so I am speaking with, as I mentioned in the intro, Sarah Fairweather, you are the Senior Program Manager of Ethics at WellSaid Labs. The job description that you have on LinkedIn says, “design and implement ethics based policies, frameworks and strategies in support of WellSaid Labs’ commitment to Responsible AI development and use in the synthetic voice industry.” Now, before, we talk a little bit about WellSaid Labs, if you wouldn't mind telling my audience a little bit about how someone who gets a B.A. in Psychology and Spanish at University of Virginia and then a Masters of Education at Loyola Marymount University ends up in such a cool job like this.
SARAH: Sure, I'd be happy to. You're right. I've had a really interesting career journey, and I would say that I focused on education and corporate responsibility and leadership development throughout my career. And along the way, I've had the opportunity to bring those experiences to multiple teams, including several at tech companies like Amazon and Code.org. And now WellSaid Labs. So I've really leaned on all of those experiences to do the work I do now.
SCOT: Very cool, I appreciate that. And so, WellSaid Labs is a synthetic voice industry tool. I think several of my listeners, maybe more, have probably heard of Eleven Labs or some other players in this space. But what is, if you could explain a little bit more about what WellSaid Labs does specifically, or, maybe what sets WellSaid Labs apart from the other players in the space and the AI voice industry, let's call it.
SARAH: Yeah, sure, I'd be happy to. So we really aim to provide our enterprise level clients with high quality, natural sounding, really trustworthy voices. And I think what sets us apart from other folks in the industry is that we were founded with ethics at our core. And we define ethics as a combination of our mission and our values and the principles of Responsible AI. So things like accountability and fairness and privacy and security, and we really lean on those to make our decisions. Our clients use our voices for things like their learning and development materials. They use it to bring their marketing content to life. We have lots of folks who are creators on YouTube, et cetera, but we really focus on the enterprise level folks who want to use these voices for positive purposes. And that's why we have such strict policies around not misusing the voices, not using it for things to incite violence, not using it for things to impersonate people, et cetera.
SCOT: Got it. And I definitely want to dig into that in a little bit. I mean, you talk about like deepfakes, misinformation, all kinds of ways people scam people. But the first thing I kind of like to get into, as I look at the WellSaid website, something that I sort of noticed that's also maybe different from other players. This is sort of like a, correct me if I'm wrong, I don't want to describe this the wrong way, but it's like a marketplace for like a catalog of voices. It's not like, I mean, other companies, people can literally like upload a sample of their own voice or any voice they've just acquired from anywhere. And then it just kind of lets them like, hey, go ahead, clone this up and start to create. But you don't, that's not part of WellSaid services, right? It's all about the professional voices that are on the platform there for use. Is that correct?
SARAH: Yes. That's actually a really great thing to call out about how we operate. Another thing that sets us apart from other folks in the industry is that we actually build our voices, based on partnerships that we have with real human voice actors. So all of the voices on our marketplace, come from data that voice actors provide to us. So we know the humans behind those voices. We have their explicit consent to build synthetic voices using their data, and we compensate them both for the data they provide us, as well as for the ongoing use of their voices by our customers. So to your point, yes, we do not allow folks to just upload, any audio data and clone the voice, because that really puts, folks at risk, I think. And that leads to the issues that we're seeing in the news now about deepfakes, you know, impersonating politicians hacking fake accounts, et cetera.
SCOT: Yeah, I mean, I think with that specific thing you're talking about there, there's so many different ways to misuse it. Yes, you can make a deepfake of a politician or a famous person, but you can also just go to anyone. Like, if you just have a grievance with someone who's not famous or not a politician, and you go to their social page and just take a sample of their voice and then make them say or do something they wouldn't say, and then it's sort of more just like a.. it doesn't have to be against a celebrity or a person of interest, I guess, is what I'm saying. You can, you know, you can use the technology, for all sorts of nefarious things. And so I think it's very interesting that, and I do see that your company is definitely taking a lot of different aspects of Responsible AI into play here. Because, for me, I get that voice cloning is an incredible technology. But to me, I've often asked myself, like, well, “why?” What are the use cases for literally, like, just going to a website and being able to upload any audio sample and clone a voice. Does everyone really need to go out there and clone their voices? I'm kind of like, “Am I missing something?” It seems to me like, a large part of the use cases are for bad actors. Not all, but, you know, it does seem that way. And then on top of that, you know, you're telling me that you pay the creators for their data used to train the models. And on top of that, I think I heard you say that also, like, if someone uses a particular actor's voice in your catalog, they're also going to get paid, let's call it a royalty or something like that as well.
SARAH: That's exactly right. Yes. So every time an actor partners with us and records a script for us, they get a fixed payment. But then we also have these quarterly, essentially royalty payments based on, the amount that their synthetic voice was used by our customers.
SCOT: Got it. I don't know if you heard of, if you saw this particular.. it's a little side note, but I saw something kind of go viral on a couple platforms. There was a woman who was hired by, I think, Microsoft for Cortana. I think Cortana is like, their deprecated assistant, kind of like an earlier AI product that they deprecated and they hired her to do, I think, the Swiss German voice. Are you familiar with this?
SARAH: I have heard this story, yeah.
SCOT: Okay. Yeah, so, I mean, it seems like to me, it's sad. I just feel like maybe she just didn't read her contract or her manager or someone just really didn't. But I guess, when Microsoft deprecated Cortana, they licensed out her voice and probably many other people that they paid to other platforms. I guess they're like, “Well, we're sunsetting this product, but let's kind of squeeze some money out of it as we close up shop.” And, you know, I guess I don't know all of the details, but I have to think that Microsoft wouldn't do that if it wasn't in the contract. They were able to do so. But the video of hers is very touching. I mean, she basically says, like, “I can't believe I sold my voice into perpetuity for $3,000.” I don't know if you had any thoughts on that.
SARAH: Yeah. So I can't speak to this specific story and all the details to your point, but I can say that this is one of the reasons why we invest quite a bit of time upfront in our voice actor recruitment and onboarding and relationship process. So, I think one thing that has really helped me in my role leading Ethics here is that I originally joined WellSaid to manage and expand our voice actor program. So that gave me a really good perspective on the needs and concerns of the humans behind our voices. I spent a lot of time in the recruitment process, talking through how we work, how the voices are used, the risks that we try to prevent with our content moderation process and program, and also the risks that we know are out there that we might not be able to detect in time, right? So we're really upfront with our actors about what we can and can't do and both the benefits and potential, cons of having a synthetic voice. We really wanted actors to be fully aware of what they were getting into before they ever signed on with us. So that's a reason why we really, again, prioritize relationships, why we put so much emphasis on consent in our work, and then why we invest so much in our trust and safety processes, like our active content moderation program, that's really aligned to our terms of service and really, calls on customers to adhere to those expectations, because there are risks in building a synthetic voice, and we want to be honest and upfront about that. And of course do our due diligence to reduce the potential harm there. But I tell all actors, like, this is an important choice you're making. We want to make sure you're fully informed before you ever sign on with us.
SCOT: Yeah, I can imagine. So you touched on a couple things here I'd like to just dive into a little bit. The first one is, sounds like when people use the platform that you have processes in place to monitor the content that's being generated, I assume using some algorithms and some AI as well as humans. But to prevent the actual, I'm sure you have terms of use, and there are things that are against the terms of use, certain things that you can't do.
And it's good to hear that you moderate for it because I think, probably plenty of companies have a terms of use, but they kind of just wait for a flag or for a user complaint. But you're telling me you actively moderate the content for misuse?
SARAH: That's exactly right, yeah. I'm glad you called that out. There are other voice companies that have terms of service, like ours, that say it's prohibited to create content that incites violence, that's discriminatory, that interferes with elections, et cetera. But we go a step further and actually actively moderate that content. You know, our systems actually scan for and block prohibited content before it can be downloaded as an audio clip. and that is, moderated both by humans as well as some automated tools. So that's a really important part of our process.
SCOT: Got it. Yeah, I mean, it's one thing to say, hey, this is against the terms, and if we hear about it, we'll cancel your account versus we're actively monitoring what's going on. So that's really awesome to hear.
The other part that I was curious about, it sounds like that you're, when you said, look, we want to be really transparent with voice actors, what they're getting into. Is there any sort of, I assume no, but maybe you could.. is there any sort of like, review process for the actor? Meaning like, it might not be against the terms of use, but if I sign a deal with you as a voice actor, not me, hypothetically, I don't want to, but then, like, I might have an issue if someone uses my voice in a project that is like, maybe particularly, left or right wing leaning, either way, depending on my political persuasion, right? Is that the sort of thing, like you might tell the actor, like, “Look, there might be something perfectly legal that's not against our terms of use that you disagree with, but it could be your voice saying those things.” Is that, that's a reality, right?
SARAH: That is the reality, yeah. And we again try to be very clear with actors that they don't have a choice about how their voice is used, right? They can't pick and choose which clients use their voice. They can count on us to uphold our terms of service and hold our customers to only the allowed use of our voices. Again, we try to be upfront with actors about that, and there have been actors who have said, “You know what, this isn't the right fit for me because I don't want my voice used in XYZ ways.” And that's okay, right? Like, we want folks to understand what this looks like and make an informed decision about whether they work with us.
SCOT: I mean, it's good that, just to have that transparency, right? Which is another hallmark of responsible AI, transparency, letting people know, look, we have terms, so your voice won't be used for hate speech, XYZ, but it could be that a customer is putting out materials one way or another on, let's say, something like climate change or something that definitely has polarization on an issue. And it might be that your voice is on the side that you don't agree with or something like that. And that's just, that's just how it goes. So as long as you're transparent about that, I think that's commendable. I think a lot of voice actors are not lawyers, so people need help. They need to have a good agent or a representation or just a lawyer that they hire.
But again, going back to this case with this woman, with Cortana, when I read it, not knowing the details, it just sort of felt like, this is really sad. But it just feels like someone didn't read the fine print, but also didn't have someone as, transparent, on the other side, perhaps just sort of saying like, “Well, it was in the doc, so I'm not liable,” which is, unfortunately, the way a lot of people work. But this is awesome. I really appreciate the context that you're giving me about your company. Is there anything we're missing here? Is there anything, anything else about, how WellSaid Labs puts their commitment of ethical use into practice that we haven't covered? We've covered a lot.
SARAH: Yeah, we have covered a lot. I think the other thing we always like to highlight is just our data practices. That's in the news quite a bit. Like concerns about where data comes from, how models are built, et cetera. And we like to stress that we build our model in house and we only build our voices using data from real humans, like I've mentioned. And then we only train our model on voice data that we have permission to use. That's really important that we do that, even if it takes more time, right? Again, it's just another demonstration of our commitment to Responsible AI. And then the other thing that's really important to all of our customers is that we don't train our model on their data, right? And clients, our customers, own the content they create on our platform.
SCOT: Got it.
SARAH: Again, I think those practices are more examples of what sets us apart from other folks in the industry.
SCOT: Got it. So if there comes a day, and I think there might, where there's sort of a.. I know people talk about a labeling system for synthetic media. I don't necessarily mean that type of label. I guess I'm sort of talking about the, the sort of fair trade, sort of rainforesty looking labels that we see in the grocery store sometimes on like coffee. But like if they were to say like, this is a fair trade model. Like I do believe there may come a time where companies like yourself, it's even like a marketing angle. Like “Look, you can feel confident that when you use our product, that the models that we trained on, we have an ethical model,” let's say. And so it sounds like WellSaid Labs would have that little frog rainforest label, right?
SARAH: Those little organic people.
SCOT: Yes, that's right.
SARAH: Yes. I believe that we would qualify for that. And there are actually a few startup organizations now that are aiming to provide that level of certification. And I believe that that's the kind of thing that customers are going to start seeking out more regularly.
SCOT: Yeah, I think so too. You know, certainly AI is changing rapidly. Everyone knows that day to day, these products, it's crazy. But I would also say that public sentiment is kind of all over the map and changing all the time too. And it really feels like certainly compared to nine or twelve months ago, there's a lot more interest in Responsible AI. I mean, I'm in Los Angeles, so the creator community here definitely has a lot to say about AI. Sometimes on the negative side. On the positive side, a lot of people are excited, but also a lot of people are not. But it's definitely, not just like this honeymoon phase of just being excited over all these crazy products being released. I think we're moving a little bit past that. So I guess this is fantastic.
I would like to, as we wrap up, just what are some other use cases? If you can't answer some of these questions, that's totally fine. But I'm asking a little bit about maybe some of the things with your company of the use cases. Can you speak to which ones are kind of most popular? Like, is it mostly corporate training videos? Or like, is it people, they're like, hey, I want to make, my own, like, animated, you know, cartoon, and I'm going to use some of the voices for my characters? Or, do you see something that kind of stands out more than the others as a popular use case?
SARAH: Oh, yeah, that's a great question. I would say, a very common use case is exactly what you said. Like, corporate training videos, learning and development materials from big corporations, everything from, you know, preparing call center employees to, you know, manage tough calls with customers to preparing healthcare organizations where, you know, how to use a new tool or how to work with patients. Everything in there. We also see some interest from curriculum providers, for incorporating our voices into their materials. And then certainly marketing teams love to use our voices for things like promotional materials, radio advertisements, et cetera. Yeah, so those are some of the big use cases.
SCOT: Cool. I want to give you the last word. I'll sort of give you a question, and then if you can't answer it, just maybe give your, you know, I want to give you a final thought, but, a prompt, maybe, that you can choose, to accept or not, is can you speak to any upcoming product features or releases that are coming to WellSaid Labs? And if that is confidential information, then maybe your final thoughts?
SARAH: Yes, I probably will not speak to what's coming up with WellSaid, but..
SCOT: Understood.
SARAH: Thank you. Other than to say that we continue to build voices that are some of the most natural sounding in the market, and we're giving our customers more, kind of directorial control over those voices, which is really interesting and is opening up more use cases. But my final thoughts, I think that I'm really trying to stress with everyone is that responsible innovation is possible, right? Even in the pressures that the tech industry faces in terms of funding and the need for speed, et cetera. I think that we're demonstrating that it's possible to be successful and responsible and I really encourage folks as they're choosing products to ask questions about how the product is built, right? Where's the data coming from? How are the humans behind that data being treated? What are the risks, to those humans and to me using this product and really make some informed choices there because I think that's going to demonstrate to leadership in the tech world that folks really do expect responsibility on the part of these leaders and the part of these companies.
SCOT: Well, no pun intended, but well said! So, Sarah, thank you so much for coming on the show. I really appreciate it.
Sarah Fairweather, Senior Program Manager of Ethics at WellSaid Labs. I really appreciate your time. Thank you so much.
SARAH: Thank you, Scot.
SCOT: All right, take care.