Speko is a routing and benchmarking platform for voice AI that connects various speech models through a unified API and provides performance data across multiple languages.
What the product does and how it is positioned
Speko operates as a centralized infrastructure layer for voice AI applications, offering a hosted router that manages data flow between different speech-to-text (STT), large language model (LLM), and text-to-speech (TTS) providers. The platform is designed to simplify the integration of multiple AI models by providing a single API and typed contracts.
A core component of the service is its benchmarking suite, which tracks the accuracy and cost of various models language by language. This allows developers to identify which models perform best for specific linguistic requirements, addressing the fact that model performance often varies significantly outside of English.
Source-supported ways to use the product
Developers can use the benchmarking matrix to select STT and TTS models that offer the best balance of Word Error Rate and cost for their target language.
The platform can be used to run and observe voice workers within LiveKit or Pipecat environments using native gateway integrations.
The documented workflow, where available
Review the language-specific benchmarks to identify the most effective STT, LLM, and TTS models for the application's requirements.
Integrate the Speko Gateway into a Python-based framework like LiveKit or Pipecat to handle provider-direct streaming.
Connect the voice agent to the hosted router using the provided API keys and configure routing objectives such as balanced performance.
Speko provides a detailed comparison of AI models across different languages to help developers navigate the variability in speech technology performance. The platform measures Word Error Rate (WER) and cost per minute for various stages of the voice AI pipeline, including speech-to-text and text-to-speech.
The benchmarking data highlights that no single model is the top performer across all languages. By visualizing these metrics, the platform enables users to make data-driven decisions when selecting providers for global applications where English-only benchmarks are insufficient.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
The Speko Router is a hosted, provider-neutral data plane that manages STT, LLM, and TTS routing through typed contracts.
The Gateway supports local Bring Your Own Key (BYOK) credentials, allowing for provider-direct streaming without sharing keys with the hosted router.
Speko provides native gateway services and integrations for the LiveKit and Pipecat frameworks, specifically for Python environments.
Yes, Speko offers a Model Context Protocol (MCP) transport that allows it to be added to tools like Claude and Cursor.
No, some models are only measured in English, and the platform notes that their rank in other languages is currently unknown.
