Casting marketplace Voice123 emailed voice actors inviting them to build AI models of their own voices, a feature open to members from its $15-a-month Starter plan. Three days later game voice actor Jonathan Fortin posted the whole email on Facebook with a scathing reply, and other performers quickly agreed with him. Voice Over Herald's reading is that the platform's consent rules are among the most careful of any major casting site, while its sales pitch badly misjudges the people it is selling to. The email opened with a statement about transparency, trust, ownership and control before getting to the offer.
The email promises five things: members can create and manage voice models, let clients generate short snippets during auditions, pick which clients get access, negotiate terms and payment, and see what clients generate. The second item is the heart of it. Called Instant Auditions in the email and Custom Script Auditions in Voice123's help pages, it lets a client type their own copy and hear it read back by the actor's model, a read the actor never gave. The actor is not in the session and never auditioned for that script, yet the client comes away with an impression of how they would sound on it.
Fortin, whose credits include the adventure game Willy Morgan and the Curse of Bone Town and this year's Whirlight, ridiculed the idea that actors would be glad to be judged on a machine's imitation of them. His argument is that the job is reading a script closely and making deliberate choices, which is exactly what a clone does not do. In his post he parodied the pitch, picturing clients hearing a poor robotic imitation making choices he would never make and deciding whether to hire him on that basis. Commenters echoed him: one member with a basic plan said she had received the same email and felt the same, another called it a corporate attempt to make actors cheer their own exploitation, and Fortin joked that he was being asked to pay monthly for less work.
He also suggested that years of audition uploads had probably fed Voice123's AI. The outlet stresses that this was his inference, and hedged as such. The company's FAQ says its model was trained on open-source licensed data and that no AI voice can be built without the actor's consent. The article's view is that both statements can stand at once and that the gap between them is the story: a clean contract does not stop a marketing email from making a decade of uploads feel exposed, and no FAQ entry repairs that kind of damage to trust.
On paper the protections are real. Voice123 says actors keep full ownership of their AI voice and it claims no rights in it, and clients cannot download generated audio until the actor accepts terms and payment. Its Voice Authenticity Initiative, launched in 2024, stamps uploaded audio with C2PA-standard credentials, and a text-to-speech policy drawn up with NAVA requires corporate client accounts, contracts in advance, payment through the platform and no perpetual buyouts, and bans clients who use samples without paying for them. Paid members can also have files matched against their stored audio. The outlet notes that most of those rules were written after abuse had already happened.
The article also lists the gaps. A model can only be deleted by emailing a support team that warns it may be slow, actors are limited to three models, and there is no floor price for AI work. On the page aimed at clients, Voice123 offers a free generator that reads up to 140 characters of a client's script in an actor's voice. The outlet notes that 140 characters is a tagline, a phone greeting or a short pre-roll, the low end of the market where many actors earn their living. The download lock still holds, so a client cannot legitimately walk away with the file, but it protects the audio rather than the actor's objection, which is about being assessed on a performance he never gave.
Actors are asked to supply a clean sample of up to a minute, and Voice123 says the quality of that sample decides how the model sounds, in one style per model and up to ten languages.
One detail struck the writer as especially odd. To show client demand, the email cites searches for Arc Raiders AI voices. Embark Studios' shooter was criticised for shipping AI-generated lines, and by March the studio had re-recorded some of them with human actors, so the demand the email advertises comes from a case many in the business saw as a failure. Chief executive Patrick Söderlund has said the studio pays actors for their booth time and separately for permission to license their voices for less important lines. The outlet suggests details like this are what make performers feel talked at.
The piece closes with practical advice: get written confirmation of any deletion, treat the uploaded sample as the product clients will judge, and set an AI rate before a request arrives. Voice actor Kabir Singh, reviewing the service in May, said he had no plans to clone his voice and that cloning should be paid at a premium, while crediting Voice123 for keeping consent in view.
For long-form, narrative and character actors a model is a shop window with little downside, the outlet concludes, because nobody swaps a long narration for a generated sample. For those living on short corporate copy, phone systems and e-learning, it competes with them on their own profile at a price they have not yet set, and they pay a monthly fee for the privilege. Voice123, it says, handled the paperwork honestly but has not yet explained to working actors why they should want this.
Sources · 1