Voice agents
Voices
Choose a text-to-speech provider and voice, and tune speed, stability, style and emotion.
Providers
| Provider | Notes |
|---|---|
| Katexs | Built-in voices with no extra key needed |
| ElevenLabs | Highly natural voices; supports speed, stability, similarity and style |
| Cartesia | Very low latency; supports emotion hints |
| Deepgram | Aura voices, fast and economical |
| OpenAI | Natural general-purpose voices |
| PlayHT, Azure, Rime | Wide language and accent coverage |
Use the play button in the Voice tab to preview any voice before you choose it. Some previews need the provider's key in your workspace.
Tuning
- Speednumber
- 0.5 to 2. Default 1.
- Stabilitynumber
- ElevenLabs only. 0 to 1. Default 0.5. Lower is more expressive, higher is more consistent.
- Similarity boostnumber
- ElevenLabs only. 0 to 1. Default 0.75.
- Stylenumber
- ElevenLabs only. 0 to 1. Default 0.
- Emotionenum
- Cartesia only: neutral, positivity, curiosity, surprise, sadness, anger.
- Chunkingboolean
- ElevenLabs only. Starts speaking before the full sentence is generated. Default on.
- Background soundenum
off(default) oroffice.
Note
Settings a provider doesn't support are ignored, not rejected. For example, the emotion setting only affects Cartesia voices.
Choosing well
- Pick a voice that matches your brand and your callers' language and accent.
- Test the voice reading numbers, dates, addresses and email addresses. These are where voices differ most.
- Prefer lower-latency providers for high-volume front desks; prefer the most natural voices for sales and premium service.
Last updated November 2026 · Found a problem? Email support@katexs.com
