Two new English voice families: natural and premium
October 28, 2026 · release
Piper is fast and cheap, and some of you told us it sounds robotic. So we added two voice families that sound closer to a person, and we kept the old default exactly where it was. Piper Amy (medium) is still what a new project speaks with. A project only changes voice when you select one.
Natural is Kitten nano: eight English voices, 30 credits per 1,000 characters. Premium is Kokoro: nine English voices, American and British, 145 credits per 1,000 characters.
The price follows the compute a voice needs, and so does the wait. We measured the first word of a short reply on our speech servers, one request at a time:
| Voice family | First word of a short reply | Credits per 1,000 characters |
|---|---|---|
| Piper (default) | about 0.2 s | 15 (medium) |
| Natural (Kitten nano) | about 0.6 s | 30 |
| Premium (Kokoro) | about 1.5 s | 145 |
A second and a half before the agent starts talking is something callers notice. Premium suits cases where the sound of the voice is the point, such as a narrated demo or a branded assistant. For fast back-and-forth, natural is the better trade, and Piper is still the quickest.
Billing works as before: characters you synthesise, at the rate of the voice's tier.
Try it
Pick a voice in the dashboard Voice picker, or from the CLI:
voicethere projects voice set --tts-model-id en-kokoro-heart
- Docs: Natural and premium voices
- Pricing: Speech pricing