Speech AI Models: TTS, STT, Speech to Speech, and LLM

One API key, one base URL, four things you can do with speech and language.

Stream natural speech with Lightning at 44.1 kHz and 200 ms to first byte.

Transcribe live or recorded audio with Pulse, 21 streaming languages.

Talk to Hydra over one WebSocket, audio in and audio out.

Run Electron, an OpenAI-compatible model built for voice agents.

The models

Premium voices, 31 languages. New checkpoint in preview for English, Spanish and Hindi.

217 voices, cloning, 12 languages.

English streaming, built-in turn detection, emotion, and gender.

Streaming and batch, 21 streaming languages.

English batch, leaderboard accuracy.

Full-duplex speech to speech.

Chat completions, 70 languages.

Numbers side by side: Compare models.

Create a key and make your first call in a minute.

Every endpoint with schemas and copyable snippets.

Plans and per-model rates are on the pricing page.