Right of publicity: employee voices and likeness in licensed recordings

What does the right of publicity protect, and does it cover voice?#

The right of publicity protects a person against unauthorized commercial use of their identity, typically their name and likeness and, in many states, their voice. It comes from state statutes and common law rather than one federal rule, so scope, remedies and whether the right continues after death all vary by state.

The classic cases involved advertising and endorsements. AI raised a newer question: whether training on someone's recorded voice, or generating a replica of it, is a use of identity. Several states have updated their laws to address digital replicas, and Tennessee's ELVIS Act, which extended that state's publicity law to cover voice, is a widely cited example. Federal proposals on digital replicas have also been introduced, so check the current status of any state or federal measure with counsel rather than assuming it applies.

For a company licensing records, the practical point is simple. Employee voices and faces are personal attributes, and whether licensing them is a commercial use of identity is assessed state by state with counsel.

Where employee voices and likeness show up in business records#

Employee voices and likeness sit in more systems than most companies expect. Sales call platforms, support phone systems, recorded video meetings, onboarding libraries and screen recordings attached to tickets all capture them.

Vendor terms are worth reading, but they cut one way. Gong's terms say it acquires no right, title or interest in customer data, and Zoom's terms say it does not use audio, video or chat customer content to train AI models. Those commitments limit what the vendor does. They do not give you any added right to license your employees' voices.

Transcripts vs audio: which format to license#

Transcripts are the default format for licensing recorded conversations because text can be de-identified and audio largely cannot. A transcript with speaker names replaced by roles such as Agent and Customer, and with names, emails and account details removed, keeps the diagnostic reasoning and problem solving that buyers value.

Audio keeps the speaker's voice, which can identify a person even with every name bleeped. Voice also enables two uses suppliers should not allow by default: building a synthetic voice that sounds like a specific person, and creating a voiceprint that recognizes them. Video adds faces, which raises the same issues for likeness.

Biometric and recording laws are separate layers#

Biometric privacy laws can apply to recordings even where publicity rights do not, if a voiceprint or face geometry is derived from them. Illinois's Biometric Information Privacy Act is the best known: a prevailing party can recover liquidated damages of $1,000 per negligent violation or $5,000 per intentional or reckless violation, and the Illinois Supreme Court held in Rosenbach v. Six Flags that a person need not show actual injury beyond the violation to sue.

Wiretap and call recording laws are a third layer. They govern whether a call could lawfully be recorded in the first place, and some states require every party's consent. Publicity rights, biometric rules and recording laws may all apply to the same audio file, and each is assessed deal by deal with counsel.

What the license should say about voices and faces#

The license should say plainly what a buyer may not do with any voice or face in the data, even when only transcripts are delivered. Transcripts can still carry a distinctive speaking style, and later deliveries sometimes add formats that were not in the first package.

Put these limits in the permitted-use section rather than in a side letter, so they travel with the data and bind any affiliate or contractor the buyer uses.

• 1No voice cloning or synthesis that imitates any person in the recordings.
• 2No speaker identification, voiceprint creation or face recognition.
• 3No attempt to re-identify speakers from transcripts or link them to other data.
• 4No use of names, images or voices in marketing, demos or model outputs.
• 5Deletion of any audio or video delivered in error, with written confirmation.

Do existing employee consents already cover licensing?#

Existing employee consents rarely cover licensing recordings to a third party for AI. A notice that calls may be recorded for quality and training purposes usually means training the company's own staff. Handbook language about monitoring company systems speaks to ownership and privacy expectations, not to licensing a person's voice or face.

Before deciding on any format, gather the paper and read it with the use in mind.

• 1Collect the actual consent documents: recording notices, offer letters, handbooks and any video or photo releases, with their dates.
• 2Check whether any release mentions third parties, AI or machine learning; most marketing releases do not.
• 3Separate current employees from former employees, whose consent is harder to obtain.
• 4Ask counsel whether new consent from current staff is appropriate and how to make it genuinely voluntary.
• 5Default to de-identified transcripts for anyone without clear, written, AI-specific consent.

Illustrative: a software company scopes its recorded calls#

Illustrative: a fictional vertical software company has years of Gong sales calls, recorded Zoom onboarding sessions and an internal video library for new support agents. A buyer is interested in how customer problems are diagnosed and resolved on calls.

The company licenses de-identified transcripts of onboarding and support calls, with speaker names replaced by roles and customer details removed. Raw audio and all video are excluded. The training video library is excluded because its presenter releases covered internal use only. The license bans voice synthesis, speaker identification and any attempt to link transcripts back to individuals, and the excluded formats are listed in the privacy record.

How SourceX handles voices and likeness#

SourceX handles recorded conversations in the Preparation step of the SourceX five-step transaction. Transcripts are de-identified, and audio or video is excluded unless the supplier holds consent that clearly covers the use. The supplier approves each format before anything is delivered.

The SourceX Evidence Packet records which formats were licensed and which were excluded, the consent documents relied on and the permitted-use limits, including any ban on cloning or speaker identification.