Speko

Speko is a routing layer for voice AI, built on a premise worth stating plainly: nearly every speech model is benchmarked in English, and most users do not speak English. Speko measures 56 speech and language models across 10 languages and routes each session to whichever model performs best for that language and workload, rather than to whichever vendor you happened to sign with.

It covers all three legs of a voice pipeline — speech-to-text, the LLM in the middle, and text-to-speech — behind a single API, so swapping a provider does not mean rewriting your integration. Routing is measured rather than heuristic, and it includes pre-response failover: if a provider degrades or times out, the session moves before the user hears a gap. There are documented integrations for LiveKit and Pipecat, the two common voice-agent frameworks.

Pricing has two self-serve paths, both in public preview, and every account starts with $100 in signup credit. Router adds 5% on top of the provider's own published rate, so you keep your existing vendor relationships and pay Speko only for the routing. Speko infra bundles all three legs into a flat $0.09 per minute all-in. An Enterprise tier locks custom rates for a contract term and adds deployment reviews, a dedicated support channel and a contractual SLA against a committed monthly minimum — the preview tiers carry no SLA.

The company is Y Combinator-backed and lists SOC 2, HIPAA and GDPR compliance badges. Voice, model and pricing tables are published openly on the site, which makes it straightforward to check the routing claims against your own language mix before committing.