What is synthetic in what we deliver
Essentially all of it, and we would rather say so plainly. The spoken voice is generated by a text-to-speech model. Music beds and sound effects are generated by models running on our own hardware. Sung jingles are generated too. The copy is written by a language model from your brief. What is not generated is the production: the arrangement, the timing, the ducking, the mastering and the quality checks are our own deterministic engine.
Why we think that is fine for imaging
Station identifications, sweepers and promos are the part of a broadcast day where a synthetic read is least likely to mislead anyone: nobody believes a sweeper is a person talking to them. We do not produce news reads, host voices, or anything presented as a real individual speaking.
What we will not do
- Clone a real person's voice without their written, verified consent.
- Generate a voice that imitates a named public figure.
- Generate anything resembling an emergency alert tone.
Disclosure on air
Some jurisdictions require, or may soon require, that synthetic audio be disclosed to listeners. Every station account can switch on an optional disclosure tag line, produced in the same language and voice, to append where local rules call for it. If you tell us where you broadcast, we will flag it when the rules in that jurisdiction change.
Provenance
Every file we produce records which model and which exact settings generated it. That record is kept for the life of the account. If a rights question is ever raised about a cut, we can say precisely how it was made and reproduce it.