---
title: Multilingual AI Agents & Custom Voices | Saloma AI
description: Choose from 70+ languages across supported AI voices. Compare models, test conversations and design an original voice with GPT-Live, subject to its limits.
canonical: https://saloma.cosinechat.com/platform/ai-agents/
updated: 2026-10-26
---

- [Home](https://saloma.cosinechat.com/)

- [Platform](https://saloma.cosinechat.com/platform/)

- AI agents & voices

Platform / AI agents & voices

# Multilingual AI agents. Your brand, your voice.

Pick the voice, language and personality that fits your business. Choose from 70+ languages across supported voice options. Test every voice in the console before your customers hear it.

Updated 26 October 2026

At a glance

- 70+ languages across voice options

- Original voice design on GPT-Live

- Voice and text tests in the console

AI voices, languages & testing

## Choose how your business sounds.

Compare AI voices and languages for your service. Test how the agent speaks and answers questions. GPT-Live also lets you describe an original voice, subject to provider access and limits.

- Choose a voice model for your needs. Options include live voice from Google, OpenAI GPT-Live and Alibaba Qwen. Another option combines Google speech recognition, a separate conversation model and ElevenLabs speech. Filter voices by language.

- Design an original adult voice on GPT-Live, Model 2. Describe its tone, pitch, pace, warmth and English accent. Provider access and quotas apply. Voice creation has no audio preview: save the voice, then listen in a voice test.

- Check model limits before choosing. Google Model 1b is Beta. Model 2 uses OpenAI GPT-Live with Turbo knowledge only. It does not support call workflows, agent-tool workflows, transfers or callbacks. The retired OpenAI Realtime model is only for existing saved agents. Qwen Model 4b is Experimental, with limited languages and Turbo knowledge only.

- Choose speech options on Model 3. Model 3a offers expressive and fast-response speech. Optional Model 3b adds ElevenLabs V4 Turbo and V4 Quality. Model 3b needs configured provider rates above zero. Model 3 lets you select the conversation model and voice separately.

- Start with ten conversation personalities. Set greetings, business facts and the local time zone. Choose one language or multilingual mode. Adjust supported speech speed, interruptions and noise controls. A regional pronunciation setting is a preference, not a guarantee.

- Follow supported language changes during a call. In multilingual mode, the agent can follow the caller’s latest substantial speech. Automatic language switching supports fewer languages than the overall 70+ voice-language range. Fixed-language agents keep their selected language.

- Test settings before applying them. Try a model, voice, language and audio setup in the console browser test. Audio meters show microphone and AI sound levels. Optional recordings and tester names help review tests. Text tests check the permitted knowledge sources, not voice or phone quality.

Features depend on the model and enabled providers. Do not use custom voice descriptions to imitate real people or include caller data. The voice provider has separate data-retention terms. Internal tests are billable. They do not run live transfers, callbacks, connected workflows or report emails. Text tests do not test audio or phone connections.

See the product

## Set up and test your AI agents.

Manage your agents, choose their language and voice, and test conversations before launch.

Real product screen with sample data. Controls depend on your role and setup.

Keep exploring

## Want to know more?

Ready when you are

## Let Saloma answer your next call.

Tell us how your calls work today. We will show you Saloma handling them, in your languages and with your tools.

Can’t wait? Tap the green call button and talk to Saloma now.

Already a customer? Sign in to manage your agents and calls.

[hey@ordinexis.com](mailto:hey@ordinexis.com)