Skip to main content
The voice configuration option allows you to choose from a variety of high-definition AI Text-to-Speech (TTS) voices for your agent. Selecting the right voice helps ensure that your AI agent communicates effectively and aligns with your brand’s identity.

​
Available Voice Options

Telzino offers voices from four TTS providers: Cartesia, Minimax, Inworld and Deepgram Flux. Every voice picker in the Dashboard and the Connect App draws from one catalogue, and each voice shows its provider beside its name. The picker groups voices by the agent’s Language — for a Spanish agent the Spanish voices come first, then everything else — and labels each voice’s language. The Language setting controls what the agent says; the accent comes from the voice, so pick a voice marked with the language you want to hear.

​
Cartesia Voices

Cartesia voices are high-quality, customer-service-focused voices optimized for phone conversations. Spanish and French Cartesia voices speak with a native accent and are listed first for agents in that language: Cartesia voices run on the Sonic 3 model by default; the version selector also offers Sonic 3.6. Sonic 2 and 3.5 are retired and fold to Sonic 3 on save.
Use the Preview Voice endpoint to hear how any Cartesia voice sounds with custom text.

​
Minimax Voices

Minimax voices offer multilingual support with natural-sounding speech. Listen to audio samples below.

​
Inworld Voices

Inworld Realtime TTS voices, with English accents from three regions. The voice ID is the full Inworld string.

​
Deepgram Flux Voices

Thirty-six English voices from Deepgram’s Flux text-to-speech, with American, British, Australian, Irish, Indian, Singaporean and Filipino accents. Voice IDs follow the pattern flux-<name>-en, for example flux-hannah-en (Hannah, American female), flux-colin-en (Colin, British male) or flux-meena-en (Meena, Indian female). Use List Voices for the full set with each voice’s description. A Partner or Reseller can lock MiniMax, Inworld or Deepgram Flux for an organization through Feature Administration; locked providers disappear from the voice pickers. Cartesia can never be locked.

​
Custom / Cloned Voices

In addition to the built-in catalogue, you can give an agent your own voice by cloning it with a short audio sample. Cloned voices are organization-level — once created, any agent in your organization can select one.
Custom Voices is an organization-level feature that must be enabled for your organization before the management UI appears. When it is disabled, agents already assigned a cloned voice keep using it, but new cloned voices cannot be created or selected.

​
Creating a cloned voice

Cloned voices are created and managed from the agents list, not from an individual agent:
  • In the dashboard, open the Custom Voices button on the organization’s row.
  • In Connect, use the Custom Voices button on the agents toolbar (it opens the Custom Voices page).
Then choose Add custom voice and either:
  • Record a sample in your browser. Recording is bounded to 10–30 seconds (it stops automatically at 30 seconds), or
  • Upload an audio file (mp3, wav, flac, ogg, or webm, up to 4 MB).
Give the voice a name, choose its language, and confirm the consent checkbox — you must have the right and the consent of the person whose voice you are cloning. The language is selectable from Cartesia’s full set of supported clone languages (English, French, German, Spanish, Portuguese, Chinese, Japanese, Hindi, Italian, Korean, Dutch, Polish, Russian, Swedish, Turkish, Tagalog, Bulgarian, Romanian, Arabic, Czech, Greek, Finnish, Croatian, Malay, Slovak, Danish, Tamil, Ukrainian, Hungarian, Norwegian, Vietnamese, Bengali, Thai, Hebrew, Georgian, Indonesian, Telugu, Gujarati, Kannada, Malayalam, Marathi, Punjabi). The chosen language is stored with the voice and applied automatically when an agent uses it, so the agent speaks the clone in that language. The clip is sent to Cartesia, and the resulting voice becomes selectable like any other Cartesia voice. The raw sample is not stored; previews are generated on demand.
Cloned voices run on the sonic-3 or sonic-3.6 model (versions v3 / v3.6), which cover all clone languages. On an agent, the voice version selector offers v3 and v3.6, and the API keeps a cloned voice_id on the sonic-3 family: an omitted voiceSettings.model counts as sonic-3, and a retired sonic-2 or sonic-3.5 is saved as sonic-3. Speed and sonic-3 pronunciation overrides work as with built‑in Cartesia voices.

​
Selecting a cloned voice on an agent

On an individual agent, you only select an existing cloned voice — it appears in the agent’s voice dropdown alongside the built-in voices. Creating and deleting clones happens in the org-level Custom Voices manager described above.

​
Managing cloned voices

In the Custom Voices manager, cloned voices appear with preview and delete controls. Deleting a voice is blocked while any agent is still using it — reassign those agents first, then delete. Deleting removes the voice from Cartesia and your organization. Cloned voices can also be managed via the API — see Create, List, and Delete.

​
How to Configure the Agent’s Voice

To select the voice for your AI agent:
  1. Navigate to the voice settings in your Telzino agent configuration.
  2. Browse the available voice options and listen to samples to choose the one that best fits your needs.
  3. Select your preferred voice and save your settings to apply the changes.

​
Custom Pronunciations

Some words — company names, place names, surnames — get mispronounced because the TTS engine has never seen them. You can teach the agent how to say them. Add the word in the agent’s Pronunciations editor, type how it should sound (for example win for Nguyen), and click Apply to turn that into IPA (International Phonetic Alphabet). Preview it, then save. Custom pronunciations are available on:
  • Inworld voices
  • Cartesia voices on the sonic-3 or sonic-3.6 model
  • MiniMax voices on English agents (MiniMax accepts English phonemes only, so the editor is hidden on Spanish and French MiniMax agents)
They are not available on Cartesia sonic-2, which cannot speak phonetic overrides. sonic-2 is retired: saving an agent that still uses it moves it to v3 (sonic-3). You enter a pronunciation once and it works on any supported voice — as long as you created it with Apply. Behind the scenes each provider is given the phonemes in the form it expects, so changing an agent’s voice between Cartesia, MiniMax and Inworld keeps its pronunciations working without re-entering them. Three cases are skipped rather than applied, with no error:
  • Phonemes typed by hand instead of generated with Apply work on Cartesia only.
  • Pronunciations created for a Spanish or French voice are ignored on MiniMax, which accepts English phonemes only.
  • Multi-word entries (for example “New York”) work on Cartesia but are ignored on MiniMax and Inworld, whose pronunciation syntax covers a single word each.
The pronunciation preview plays through the agent’s own voice, so what you hear when you preview is what callers will hear.

​
Tips for Choosing the Right Voice

  • Match Your Brand: Choose a voice that reflects your brand’s personality and tone. For example, an energetic and engaging voice might be suitable for a casual eatery, while a calm and authoritative voice could be ideal for a professional service.
  • Consider Your Audience: Think about what your customers might prefer. A clear and pleasant voice is generally easier to understand and can be more enjoyable for longer interactions.
  • Test Different Voices: Don’t hesitate to try out several voices to see which one resonates best with your customers and suits various scenarios.

​
Important Note

Certain AI TTS voices may not be available when Spanish is selected as the language. Ensure to review the available options under the chosen language settings. By selecting the appropriate voice for your AI agent, you can enhance customer interactions and ensure that your communication is both effective and aligned with your brand’s identity.