Speko (YC S26)
OpenRouter for Voice AI. Every voice model is benchmarked in English. Your users don't speak English. Speko measures 83 speech and language models across 17 languages, then routes each session to the one that actually wins. I'm Bek, founder of Speko, a platform that finds an optimal combination of speech-to-text, LLM, and text-to-speech models, given your constraints, among all our public benchmarked options, and tells you why.
Reactions (1)
Similar products
by description similarity
Moonshine Open-Weights STT models
github.comhigher accuracy than WhisperLargev3
Addressee detection for voice agents: device-directed speech detection that runs before STT, so background speech, side conversations, and the agent's own TTS echo never trigger it. No wake word, model-agnostic, drop-in for LiveKit, Pipecat, ElevenLabs, Twilio, and OpenAI. The layer your VAD and turn detection are missing.
Voice gender classifier for European voice AI (1MB
huggingface.coONNX, 4ms)
Over the past few months we've been working on building a platform to build, scale and monitor voice based LLM applications.
LLM Round‑Trip Translation Benchmark
github.comWhen a model translates out of English and then back to English, how much meaning and voice does it keep? Each model does both steps (English → target language → English).