Voices and languages
Build globally-aware AI agents with native-level fluency across 90+ languages and 1000+ realistic voices.
Build globally-aware AI agents. Our platform provides native-level fluency across 90+ languages with highly emotive voice generation.
- 90+ Languages
- 1000+ Realistic Voices
- Sub-200ms Latency
Core capabilities
Our platform combines Speech-to-Text (STT) and Text-to-Speech (TTS) so your agent can hear and speak naturally.
Language Understanding (STT)
Your agent hears and transcribes words with precision. We use high-fidelity models that excel at recognizing intent across diverse regional accents. Real-time transcription, dialect and accent support, automatic language detection, telephony-optimized models.
Speech Generation (TTS)
Your agent speaks using realistic, brand-aligned voices. Access a library of 1000+ personas designed for everything from support to storytelling. Emotive and natural pacing, 1000+ voice library, multilingual v2 support.
Recommended settings
Starting points that work well for most agents. Tune from here after you test with a real script.
- Text-to-speech (TTS): Cartesia is a good default for low latency and natural voices. To speak in your own brand voice, see Voice cloning.
- Speech-to-text (STT): Soniox for Indian and global languages, or Sarvam for Indian languages. Turn on the noise reducer, and set the silence timeout to around 200 ms (how long the agent waits after the caller stops speaking).
- LLM: GPT-4.1 Mini at temperature 0.35 with streaming on. Streaming lets the agent start speaking while the reply is still generating, which lowers latency.
- Minimum words to interrupt: default is 2. Set it to 3 so short words like "yeah" or "okay" do not cut the agent off.
Enable only the languages the agent needs. Extra languages add processing.
Global support matrix
Comprehensive coverage for both understanding and speaking across 90+ locales.
Indian languages
Global languages
Multi-lingual switching. Your agent can automatically detect the language being spoken and respond in kind, so the conversation flows in whatever language the caller uses.