Voice catalogue: an inventory you can choose from
A voice catalogue is an inventory the platform supplies. Five entries have one. Only one publishes it with enough structure to filter before committing, and the rest assert it or hide it behind an endpoint. As of 2026-09-12.
| Model | How the catalogue is published | What a production can do with it |
|---|---|---|
| D-ID | Delegated to five named providers | Audit five upstream lists |
| Hedra | Behind a voices endpoint | Enumerate it in code |
| PixVerse | Built-in voices, named as a category | Nothing before a call |
| SceneMixer | A preset library, named in a checklist | Nothing before a call |
| Synthesia | Per voice, with a language code and an id | Shortlist by market and gender |
Inclusion rule. Entries whose documentation says a production may choose a voice the vendor supplies. An entry that only accepts a recording does not earn a row. Order. Alphabetical by model name.
1A list is not the same as a browsable inventory
Four of these five assert a catalogue and one publishes it. The published one carries a formal and native language name, a code, a gender, a name and an id per voice, which is enough to filter automatically and enough to audit later.
The others hand the enumeration to an endpoint or to somebody else's documentation. Both are reasonable engineering and neither helps a producer deciding whether a series can be cast at all.
2A catalogue behind an endpoint is current and unreadable
An inventory retrieved at runtime cannot go stale against the code, which is the argument for it. It also cannot be read by anybody evaluating the platform, because evaluation happens before there is any code.
So the same design is strong for the engineer and useless for the commissioner. This register records the documentation because that is what the second reader has.
3Delegation is more candid than a vague claim
Naming five outside speech providers tells a reader exactly where the real catalogue lives, which is better documentation than a sentence asserting broad support. It also imports five sets of licence terms and deprecation schedules.
Recording the provider alongside the voice id in a production's own notes is the cheap defence. The returned file does not say which service spoke, and a year later that is the one thing needed to reproduce a character.
4Sources read for this entry
This page defines the term and logs what each vendor documents about it, each figure read on 2026-09-12. Related: Language code, Guide track.