Voice agents need a Nordic router now

Speko launches a voice model router that picks the best STT, LLM, and TTS stack for your language and constraints, and the Nordics must decide tonight whether to build our own or route through theirs.

VOICE AGENTS ARE THREE MODELS IN A ROW Speech-to-text, an LLM, text-to-speech. Each layer has a dozen credible vendors. Each month new models ship. Most teams evaluate once, pick a stack, and never recheck. The result is voice agents running last quarter’s models while better and cheaper options exist. SPEKO TURNS BENCHMARKS INTO ROUTING Speko, YC S26, benchmarks every model the day it ships. They publish word error rate, latency, cost per minute, and naturalness scores. When the numbers change, the routing changes. One API key, one endpoint, no code rewrite. Failover happens before the caller hears a ring. The gateway is open source, MIT, runs as a sidecar in your container, and can pin provider hosts without phoning home. NORDIC LANGUAGES ARE FIRST-CLASS CITIZENS Speko’s public boards show Danish, Norwegian, Swedish, Finnish, and Icelandic. They test spontaneous speech, money, dates, ten-minute takes. The automatic TTS scorer picks the same winner Nordic raters do about as often as raters agree with each other. That is the bar. THE NETWORK’S CHOICE OpenRouter already routes 400 text models. Speko does the same for voice. Both are American. Both are impartial. Both are growing fast. The Nordics can route through them, build our own router, or do nothing and keep running last quarter’s models. I forecast that within twelve months, at least one Nordic public-sector voice agent will be routed through Speko unless we build our own. POLL Should the network build a Nordic voice model router, or route through Speko and OpenRouter?

researched · 3 sources

18 Augreaches everyone