Skip to main content
Your agent needs a voice and language to communicate with customers. Here you’ll pick from thousands of professional voices, preview them, and configure the AI model that powers the speech generation.

​
Steps

1

Open the voice settings

In the agent editor, scroll to the Agent voice section and click the row showing your current voice (or “Select a voice” if you haven’t picked one yet). A panel opens on the right side.
2

Search and filter the voice library

The library shows two categories:
  • Premade — The most popular voices (Rachel, Sarah, Adam, etc.), always available.
  • Community library — 5000+ voices contributed by users, with rich metadata.
Use the search bar to find a voice by name. Refine results with these filters:
  • Quality tier — Choose “Professional,” “High quality,” “Famous,” or “Any quality” (default).
  • Language — Filter voices that support a specific language. Defaults to “Any language.”
  • Gender — Show only Female, Male, Neutral, or all genders.
Tap “Clear” to reset all filters at once.
3

Preview a voice

Click the play button next to any voice to hear a sample. Click again to pause. Only voices with a preview URL can be played.
4

Pick your voice

Click on a voice name (or its metadata row) to select it. A checkmark appears. The panel closes and your voice is now set.
5

Choose your TTS model

Below the voice picker, select the TTS model family:
  • V3 Conversational — Highest quality and most expressive. Best for engaging conversations.
  • Flash — Lowest latency; responds the fastest. Best for snappy back-and-forth.
  • Multilingual — Balanced quality across 29 languages. Single model handles your language list.
The model you pick affects how well your voice adapts to different languages and the fine-tuning options available.
6

Configure stability, speed, and similarity (optional)

If you selected Flash or Multilingual, you’ll see three sliders:
  • Stability — “More expressive” (0) lets the voice vary its tone; “More consistent” (1) keeps it steady.
  • Speed — Adjust playback speed from slower (0.7x) to faster (1.2x).
  • Similarity — “Low” (0) uses a more generic timbre; “High” (1) closely matches the original voice.
(V3 Conversational doesn’t expose these; use Expressive mode instead to add emotional inflection.)
7

Set primary and additional languages

Below the voice settings, configure which languages your agent speaks:
  • Click your primary language row to pick the main language. This is where customers will start conversations.
  • Tap + Add additional languages to enable extra languages. The agent can switch between them based on customer input or context.
Languages appear with their flag emoji for quick recognition—for example, “Norwegian” or “Spanish.” You can remove any additional language by clicking the X button.
Preview URL and category tips
  • If a voice doesn’t have a preview URL, the play button is disabled.
  • The “Premade” voices are the most stable and widely used; they’re a good starting point if you’re unsure.
  • Community voices are curated and tagged with accent, gender, age, and use case (e.g., “friendly,” “narrator”) to help you find the right fit.
Model and language compatibility
  • Not all models support all 29 languages equally. The Multilingual family is designed for broad language coverage; V3 Conversational excels at a smaller set.
  • Check your agent’s voice settings after changing the model family, especially if you use multiple languages—quality may vary per language/model pair.