Written from 15 named sources · Oct 16 · first result A research companion to VO Infusion Episode 11, "Cloning - The Ghost in the Booth." Researched with Pig Knuckle and reviewed by Deb Munro and Keith Norton, with sources for every claim. Key Findings Modern text-to-speech systems generate new speech from text rather than splicing together recorded words. The model learns the statistical markers of a voice, such as pronunciation patterns, pacing, timbre, age and accent, and then produces sentences the person never recorded [6]. Instant voice cloning can build a working replica from as little as one to a few minutes of audio, while professional cloning requires at least 30 minutes of high-quality source recordings, with several hours delivering the best results [8]. ElevenLabs' Voice Library hosts roughly 10,000 community voices and has paid about $10 million to creators, but the supplied research does not include a verified current rate card, payout formula or audit mechanism for voice owners [7]. In Lehrman v. Lovo, Inc., decided July 10, 2025, the court allowed New York right-of-publicity, consumer-protection and breach-of-contract claims to proceed while dismissing the federal trademark claims and most copyright claims [1][2][4]. Lovo filed for Chapter 7 bankruptcy on May 27, 2026, automatically staying the case, which means an early courtroom win is no guarantee of a payout [4]. ACTRA Toronto's guidance centers on consent, compensation and control, and NAVA's #fAIrVoices campaign uses the same three words; ACTRA's rider is suggested language that has not been officially negotiated [15]. Canada has no harmonized federal right of publicity, and the provincial patchwork leaves real gaps: some provincial statutes expressly cover voice, British Columbia's does not, and Quebec relies on Civil Code protections [12]. How Voice Clones Are Made Today When Deb and I broke down the tech on Episode 11, we started from a simple reality: our recorded voices can end up working in places we never booked. Understanding how that happens means understanding what changed under the hood [6]. A voice clone is a model that learns the statistical patterns in a person's recorded speech and uses those patterns to generate new audio [6]. That is the practical distinction that matters to a performer. The system is not replaying a shelf of recorded words; it has learned enough about how you sound that a customer can generate new lines without booking another session [6]. Older digital systems took a different path, converting text into phonemes and then processing those into waveforms as a signal-processing problem [7]. Modern systems treat synthesis as an understanding problem: the model learns context, emotion and character from large amounts of speech, then generates audio that reflects what the text means [7]. The architecture details, things like neural vocoders and acoustic representations, are standard practice in the field, but the sources supplied for this piece do not document those specifics, so treat the deep technical layer as general industry knowledge rather than something cited here. The commercial tiers are easier to verify. ElevenLabs, the platform we focused on in the episode, offers two cloning paths [8]: Dimension Instant Voice Cloning Professional Voice Cloning Input audio As little as one minute; a few minutes of casual audio [8] At least 30 minutes of high-quality recordings; several hours for best results [8] Availability From the Starter plan ($6/month as of August 2026) [8] From the Creator plan ($11/month as of August 2026) [8] Quality Noticeably lower; fine for quick projects [8] More realistic and consistent; suited to long-form production [8] Consent checks None documented in the supplied research Requires proof of consent from the voice owner [7] That last row deserves your attention. ElevenLabs implemented verification requirements for Professional Voice Cloning after early misuse of the platform, meaning users must provide proof of consent from the voice owner [7]. It is a real safeguard, and it is also a safeguard that only reaches people who follow the rules. The practical takeaway for a working voice actor: instant cloning from a few minutes of audio is real, cheap and available to anyone with a subscription [8]. Your commercial demo, a podcast interview, an explainer on your website... any of that can be reference material. Professional cloning produces a much better replica, but it needs clean, directed studio material, which is exactly the material a scam audition can trick you into providing [8]. How Voice Marketplaces Pay and What Owners Can Control ElevenLabs describes the Voice Library as a marketplace where anyone can create an AI voice clone, share it with the community, and earn money when other users generate with it [7]. By early 2026 the library hosted nearly 10,000 voices, and the company reported paying $10 million back to voice creators [7]. That is the opportunity side, and it is real: a licensed clone can handle revisions, localization, accessibility work and low-budget narration without you attending every session [7]. The same platform also runs a licensed music model and enterprise compliance features like zero-retention data modes, which shows the tool can be operated responsibly [7]. Now the part I could not verify, and I want to be straight with you about it. The supplied research does not contain a current Voice Library rate card, payout formula, minimum payout threshold or audit mechanism for voice owners [7]. I cannot tell you whether creators are paid per character, per generated minute, per subscriber use or from a fixed pool, because no source in this research packet documents it. Pricing elsewhere on the platform also shifts: one review reports the Creator plan at 121,000 credits per month as of August 2026 [8], while another reports the allowance expanded to roughly 440,000 characters entering 2026 [9]. So treat any figure you see as "as of" a date, and check the live terms yourself. What the record does support is the structural caution. A marketplace listing means your voice is available to subscribers you did not choose, generating content you cannot pre-approve [7]. The platform's own safety controls, consent verification for PVC, watermarking and abuse reporting, exist [7], but the supplied research does not document per-use approval rights, exit clauses or model-deletion guarantees for library contributors. Before enrolling, obtain the live terms directly from the platform, save a dated copy, and read who owns the recordings, whether outputs can be sublicensed, whether the license survives account closure, and how you would remove the voice from active use. Recurring payments sound appealing. Recurring income without documented usage controls is a trade you should make deliberately, not by accident. What the Lovo Case Shows About Unauthorized Cloning Here is the documented case, and it is closer to home than any of us would like. Paul Lehrman and Linnea Sage, professional voice actors working in New York, sued Lovo, Inc., alleging the company used AI to synthesize and sell unauthorized clones of their voices [1][2]. The suit was filed May 16, 2024, in the Southern District of New York [4]. The allegations are worth reading carefully, because they show how a commercial workflow can be abused. Lehrman says he was hired through Fiverr in 2020 by an anonymous user, later identified by Lovo's counsel as a Lovo employee, to record scripts for "research purposes," with assurances the recordings would not be repurposed; he was paid $1,200 [2]. Sage says a Fiverr user who plaintiffs allege was Lovo's co-founder told her the scripts were test scripts for radio ads that would not be broadcast; she was paid $400 [2]. Both say they later discovered their voices marketed on Lovo's Genny platform, Lehrman's under the stage name "Kyle Snow" [2]. I will keep saying "allege" and "say," because these are allegations the court presumed true for procedural purposes, not findings of fact after trial [2]. On July 10, 2025, Judge J. Paul Oetken ruled on Lovo's motion to dismiss [2]. Here is how the claims sorted out: The state-law claims survived. The court held that claims for misappropriation of a voice may be properly asserted under Sections 50 and 51 of the New York Civil Rights Law, which protect against unauthorized commercial use of a person's identity, including a voice, and it allowed the consumer-protection and breach-of-contract claims to proceed [1][2]. The court also rejected Lovo's Statute of Frauds defense at this stage, noting that electronic communications may satisfy New York's writing requirements [5]. The federal claims mostly did not survive. The Lanham Act claims were dismissed because the voices, as used, did not qualify as distinctive marks or source identifiers [2][5][6]. Most copyright claims were dismissed, with the training-based copyright claims dismissed with leave to amend [1][6]. The reasoning that matters to us: copyright protects the recordings, not the voice itself [6]. Then the case took a turn that should sober every performer. On May 27, 2026, Lovo filed for Chapter 7 bankruptcy, and the case was automatically stayed [4]. Winning an early procedural victory does not mean collecting damages. If an AI startup burns its capital, you may end up in line behind other creditors. That status comes from a case tracker rather than the court opinion, so check the federal docket before relying on it for any real decision [4]. For Canadian listeners, the picture is patchier. Canada has no harmonized federal statutory right of publicity [12]. Provincial privacy statutes in Manitoba, Saskatchewan and Newfoundland and Labrador expressly cover voice; British Columbia's statute protects name and portrait for advertising but does not expressly mention voice [12]. Ontario recognizes misappropriation of personality at common law, but the tort generally requires commercial gain or endorsement-type use, which creates friction when a platform quietly absorbs a voice without an obvious ad campaign [12]. Quebec's Civil Code offers its own privacy protections [12]. No Canadian case law on voice cloning exists yet in the record supplied here, so treat that analysis as a map of the terrain, not a route that has been walked. Finding and Removing an Unauthorized Clone Start with an uncomfortable technical truth. Because modern systems generate new audio rather than copying your files, standard audio fingerprinting will not catch a clone the way it catches a pirated master [6]. Your protection is process, not automation. A monitoring workflow you can actually maintain: Search deliberately, monthly. Run your name, stage name, demo titles and distinctive phrases in quotes, plus combinations like your name with "AI voice" or "voice clone." Repeat in Google and inside the major voice platforms, because marketplace listings may not surface in ordinary search results. Test only where authorized. If a service has a public preview, compare the suspected voice against your demos and preserve the URL, account name, screenshots, sample output, date and the exact listing language. Do not violate a site's terms just to build a case. Keep an evidence folder. Original demo files, contracts, invoices, registration records and a dated log of what you found. The goal is showing identity, access and unauthorized use without editing anything in a way that undermines authentication. Report through formal channels. Use the platform's copyright, impersonation, privacy or abuse process, and ask for the listing, model and generated files to be disabled pending review. If the use involves false endorsement, fraud or explicit material, name that harm precisely rather than calling it all "AI." If the platform ignores you, escalate upstream: the hosting service, the app store, the payment processor. In the U.S., your legal theories depend on state publicity law, copyright in the underlying recordings and the facts of the use, and Lehrman shows those theories do not all get equal treatment [1][2]. California amended Civil Code Section 3344 in 2025 to add an expedited injunctive-removal mechanism, but the legislature vetoed an amendment that would have expressly extended the statute to digital replicas, so whether Section 3344 reaches AI voice clones remains unsettled as of early 2026 [13]. Tennessee's ELVIS Act, effective July 1, 2024, is clearer: it expressly protects voice, including simulations, living or dead [14][15]. In Canada, get province-specific advice, because the tort framework is incomplete rather than uniform [12]. And say it with me: before sending threats or signing any settlement a platform offers you, talk to an entertainment lawyer or your union. That is not a formality. It is the difference between a demand letter and a defamation counterclaim. What Unions and Performer Organizations Recommend ACTRA Toronto's guidance is built on three principles: consent, compensation and control over the use of a performer's image or voice [15]. NAVA's #fAIrVoices campaign uses the same three words, and it includes a pledge from five prominent online casting companies never to use voice actors' audio for AI training without consent [15]. ACTRA Toronto is a proud supporter of both [15]. Consent. Performers should have the right to consent to, and be credited for, the use of their name, image and likeness rights in new works and in AI model training [15]. SAG-AFTRA's 2023 TV/Theatrical Agreement generally requires informed consent, plus compensation in most cases, before a producer creates or uses a digital replica [11]. The 2025 Interactive Media Agreement tightened that to separate, written, specific consent for vocal or visual digital replicas, and added the right to revoke AI consent during a strike [11]. The 2026 TV/Theatrical Agreement followed, requiring "significant additional value" before a synthetic performer can replace a human role [11]. Those protections are contract-specific: a non-union job, an audiobook deal and a platform upload each run under different rules, and a union clause does not automatically follow your voice into every later use. Compensation. ACTRA's position is that performers should be compensated for all AI uses of their NIL rights, their voice included [15]. ACTRA also flags the core economic problem: one-off payments often fail to reflect that a voice may be used in perpetuity across thousands of projects [15]. SAG-AFTRA's framework ties replica use to negotiated consent-plus-compensation rather than a blanket buyout [11]. Note that NAVA's specific rider terms are not in the supplied research, so I cannot quote their model contract language here; check NAVA's current resources directly. Control. Once a digital replica exists, ACTRA argues any company handling that data must commit to safe storage and tracking of the files [15]. That is the standard to negotiate for: script review rights, term limits, and deletion with written certification when a license ends. One caution on sourcing: ACTRA's Commercial AI/Digital Replica Rider is sample language for use under the National Commercial Agreement, and ACTRA states plainly that it has not been officially negotiated, so an engager may reject or revise it [15]. Bring it as your opening position, not as a fait accompli. What This Means for Your Voice Business Run a voice-rights inventory this month. List every demo, audition, coaching sample and public interview online, with the URL and who commissioned it. Mark which files carry contractual AI restrictions and which were posted without any written AI clause. Attach an AI rider to every new quote. Separate language covering model creation, training, generated speech, term, territory, media, exclusivity, compensation, approvals and deletion. ACTRA's rider is a workable template [15]. If a client balks at "no AI training without separate written consent and negotiated compensation," that tells you something about the client. Guard your audition audio. The Lovo allegations describe talent being sourced through small freelance gigs with vague explanations [2][3]. Keep custom auditions short, ask what happens to the files, and think twice before delivering unusually long, phonetically diverse reads to an anonymous requester. Search monthly and log what you find. Your name, your phrases, the major voice platforms, in the platform search tools as well as Google. A repeatable log beats one panicked search after something appears. Price control, not just minutes. If a client wants a clone, quote the model license and the permitted uses as separate rights, with renewal, localization, exclusivity and termination terms in writing. A permanent buyout by default is a bad default. Escalate to humans when the stakes rise. A suspected unauthorized commercial clone, an exclusive license, a political or explicit use, or a demand to sign away future rights deserves review by a qualified lawyer, agent or union representative before you sign anything or send anything. Go Further The Lehrman v. Lovo opinion (790 F. Supp. 3d 348, S.D.N.Y. 2025): the primary decision, showing exactly which claims survived and which did not, and why state publicity law carried the day [1][2]. ACTRA Toronto's AI resources: the most directly useful supplied source for Canadian performers, with a best-practices guide and sample rider language [15]. SAG-AFTRA agreement coverage: a sourced walkthrough of the 2023, 2025 and 2026 contracts' AI terms, including the documented gaps critics keep naming [11]. The Canadian Bar Association essay on voice cloning: a clear explanation of why Canadian privacy torts leave voice-cloned artists underprotected, province by province [12]. The U.S. Copyright Office material summarized in the Lovo coverage: useful background on why copyright in a recording and control over a person's vocal identity are different legal questions entirely [3]. About VO Infusion VO Infusion is a podcast about artificial intelligence and the voice acting industry, co-hosted by voice actor and coach Deb Munro and voice actor Keith Norton. Episodes cover voice cloning and licensing, legal protections for performers, prompt engineering for voice actors, AI tools for the business side of voiceover and why human performance still matters. Listen wherever you get podcasts or on YouTube at @VOInfusion, and find more at voinfusion.com. About Pig Knuckle Pig Knuckle (pigknuckle.ai) is a research orchestration platform from Alchemy Agentic, founded by Paul Langtry and Keith Norton. It breaks a research question into steps, sends each step to the AI model best suited to it, and builds its citation list from the documents it actually read, so every claim can be traced to a source. Sources [1] Lehrman et al v. Lovo, Inc., 1:2024cv03770 - Document 45 (S.D.N.Y. 2025) — https://law.justia.com/cases/federal/district-courts/new-york/nysdce/1:2024cv03770/621575/45 [2] Lehrman v. Lovo, Inc. (Lehrman v. Lovo, Inc., 790 F.Supp.3d 348 (S.D. N.Y. 2025)) - vLex United States — https://case-law.vlex.com/vid/lehrman-v-lovo-inc-1098786772 [3] Lehrman v. Lovo, Inc.: Voice Actors Take on AI Voice Generation Art Law — https://www.hhrartlaw.com/2025/01/lehrman-v-lovo-inc-voice-actors-take-on-ai-voice-generation [4] Lehrman v. Lovo: Case Status AI Lawsuit Tracker — https://ailawsuittracker.com/cases/lehrman-v-lovo-1-24-cv-03770-jpo [5] Lehrman and Sage v. Lovo, Inc.: A Landmark US Decision on Voice Cloning and AI — https://www.ddg.fr/actualite/lehrman-and-sage-v-lovo-inc-a-landmark-us-decision-on-voice-cloning-and-ai [6] The Law Speaks Up: AI Voice Cloning and Consent – Juris Magazine — https://sites.law.duq.edu/juris/2025/11/25/the-law-speaks-up-ai-voice-cloning-and-consent [7] What is ElevenLabs? History of AI Voice & TTS (2026) — https://www.taskade.com/blog/elevenlabs-history [8] ElevenLabs Review 2026: Voices, Cloning, Dubbing, API & MCP — https://delavo.de/en/elevenlabs-ai-voices [9] ElevenLabs Review 2026: Overage, Billing Cap & Creator Plan — https://belreos.com/tools/elevenlabs [11] Three Years of SAG-AFTRA's AI Protections: What's Actually Changed, VoiceEditSuite — https://voiceeditsuite.com/blog/sag-aftra-ai-agreement-three-years-later [12] “Fake Drake” and the Real Harm: Canadian Privacy Torts Fail to Protect Voice-Cloned Artists Canadian Bar Association — https://cba.org/sections/privacy-and-access/resources/essaywinnerpriv2025 [13] Defamation & Reputation Management 2026 - USA – California Global Practice Guides Chambers and Partners — https://practiceguides.chambers.com/practice-guides/defamation-reputation-management-2026/usa-california/trends-and-developments [14] What Is Tennessee’s ELVIS Act? A New AI Likeness Law — https://www.identity.org/what-is-tennessees-elvis-act [15] Artificial Intelligence Resources - ACTRA Toronto — https://actratoronto.com/artificial-intelligence-resources AI and Voice Actors: Statistics for 2026 (US and Canada) — https://flauntaudio.com/ai-voice-actor-statistics-2026