Custom wake words: what voice platforms actually let you change

Brand owners routinely arrive wanting their character to answer to a custom name, and are surprised that the answer depends on which voice platform the product uses. Wake-word and voice customization is a platform-terms question, and it belongs in the platform-selection meeting, not the firmware backlog.

The three layers of voice identity

The wake word — what the toy answers to. The speaking voice — the synthesized or recorded persona. The phrase engine — what the toy can say and when. Each layer has different openness per platform, and the differences are contractual, not technical.

What is typically open

Offline-platform wake words are usually the most flexible: engines exist that train a custom wake phrase from a handful of samples, and branded names train well when they are phonetically distinctive. Recorded phrase libraries are open almost everywhere — your character speaks your recorded lines, so long as licensing is sorted. Speaking voices for cloud TTS are platform-owned; you often choose from a catalogue rather than commissioning a voice.

What is typically closed

The big cloud voice platforms’ wake words and assistant identities are tightly held — a toy can work with them but rarely rebrands them. And some platforms’ wake-word model training is gated behind volume commitments or certification programs. Neither is a problem — it is a term sheet. It becomes a problem only when discovered after tooling.

How we put this in the proposal

Voice identity gets a line in every platform comparison we write: what can carry the brand’s name, what can carry the brand’s voice, and what is fixed. Clients pick a platform knowing which parts of the character’s identity belong to them and which are borrowed.