Speko

Speko is one API for the full voice stack: speech-to-text, LLM and text-to-speech, routed to the measured-best provider by public benchmarks. Most voice AI platforms ask you to take their word for it. We publish the measurements. Our open benchmark at benchmarks.speko.ai tests the major STT, LLM and TTS providers the same way, continuously - accuracy, latency and cost, per language and use case. Each call runs on the stack that measured best.

We also run a blind listening test at arena.speko.ai: real listeners pick which AI voice sounds more human, ranked against a real human baseline with Bradley-Terry statistics. Describe the agent you want in a sentence. Test it in the browser or on a real phone call. Ship it to a phone number, a web widget or the API. What developers get: - TypeScript and Python SDKs - A hosted MCP server (mcp.speko.ai), so coding agents can build and operate voice agents - A LiveKit adapter - A playground at speko.ai/build, no account required - Sign-up to your first transcribe call in under five minutes Your audio and transcripts are not used to train models.

SOC 2 in progress. Backed by Y Combinator (S26). Site: https://speko.ai Docs: https://docs.speko.dev Start free: https://platform.speko.dev