Speko routes speech-to-text, LLM, and text-to-speech models through one API. It benchmarks models by language so teams can pick a stack based on their constraints.