Contracts for American voice work have expanded substantially, and the added length is almost entirely about what may be done with the recording after the job ends.
A recording became an input as well as a product
Historically a voice session produced a deliverable used in a defined context, and the contract described that context: this commercial, this market, this term.
The same recording can now serve as training material for a synthetic voice capable of saying things the performer never said, for as long as the model exists.
That is a categorically different use, and existing contract language written for broadcast rights did not address it in either direction.
Broad grants stopped being acceptable
Standard agreements often included expansive language covering media now known or later devised. That phrasing was written to future-proof distribution formats.
Read against synthetic voice, it appeared to permit training and generation, which is not what performers understood themselves to be agreeing to when they signed.
Negotiators on the performer side responded by demanding that synthetic use be named explicitly rather than left to a general clause, which is why the documents grew.
Consent is now specified per use
Current agreements tend to break the grant into separate permissions: use of the recording as delivered, use for training, and generation of new performances from the trained voice.
Each carries its own compensation and term, on the reasoning that agreeing to one does not imply agreeing to the others.
The separation also allows a performer to accept the original job while declining the training use, an option that a single combined grant did not offer.
Duration and revocation are the difficult clauses
A conventional usage term expires, after which the material stops being used. A trained model does not naturally expire, and removing a contribution from it is technically awkward.
Agreements have attempted to address this with commitments to stop generating after a term, to delete the trained artifact, or to seek renewal.
Those commitments are only as good as the ability to verify them, which remains the least settled part of the arrangement.
Estates and deceased performers required separate language
A synthetic voice can outlive the person, which raises who may authorize its use afterward and on what terms.
Several US states have post-mortem rights covering name, likeness and voice, but the duration and scope differ enough that a national contract cannot rely on one rule.
Agreements increasingly name the party who controls post-mortem authorization explicitly, because leaving it to the applicable statute produces different answers depending on where the performer lived.