Speko

Speko is a voice-AI infrastructure company that builds a provider-neutral routing layer for speech and language models. Its core proposition is that no single speech model is best across every language and every stage of a voice pipeline, so applications should be able to route each request to whichever provider and model best fits the objective. Speko exposes this through a hosted data plane, Speko Router, at router.speko.dev, which offers typed operations for model discovery, speech transcription, speech synthesis, and LLM responses. Teams can let Router choose an eligible route by latency, quality, cost, or a balanced objective, or pin a specific provider and model; each response reports the selected route in structured metadata. A second product boundary, Speko Gateway, is an open customer-side runtime that runs beside LiveKit or Pipecat, keeps bring-your-own-key provider credentials in the customer process, and exposes one local streaming protocol across providers. Gateway can use provider-direct routes, Speko-managed routes, or both, and its public repository owns the canonical wire contract consumed by the hosted Router. A separate hosted Agents API at api.speko.dev owns agents, sessions, phone calls, and telephony. Speko describes these surfaces as complementary but not interchangeable. Speko markets itself on benchmark coverage by language. Its site presents speech-to-text and text-to-speech comparisons across languages including English, Arabic, French, German, Hindi, Norwegian, Spanish, Tamil, and Telugu, drawing on models from providers such as Deepgram, Soniox, OpenAI, Alibaba, Cartesia, Google, Smallest, Gradium, AssemblyAI, ElevenLabs, Gladia, Inworld, Modulate, xAI, and Meta. The company highlights that many models are measured only in English and that different models win across different languages, which is the core argument for routing rather than a single default model. Speko is operated by Speko Group, Inc. from the United States and is a Y Combinator company. Product, support, security, billing, and partnership inquiries reach the team at team@speko.ai. The company positions itself for two kinds of buyers: applications that want Speko to operate the managed STT, LLM, and TTS data plane, and teams that own their own conversation loop and want provider-direct streaming or local credential custody.

Speko emerged as a voice-AI infrastructure company and is described in its own materials as a Y Combinator company operated by Speko Group, Inc. from the United States. Its founding year and founders are not stated in the available evidence, so they are not asserted here. The company's product strategy centers on a single observation: the voice-AI stack is fragmented across many providers, and no single model is best across every language or every stage of a voice pipeline. Speko's site states that 12 of 24 benchmarked models are only measured in English, that their rank in any other language is unknown, and that four different models win across nine languages. This is the argument for a routing layer rather than a fixed default model. Speko built two distinct product boundaries around that argument. Speko Router is the hosted, provider-neutral data plane at router.speko.dev. It exposes typed operations for model discovery, speech transcription, speech synthesis, and LLM responses. Teams can let Router choose an eligible route by latency, quality, cost, or a balanced objective, or pin a provider and model, and each response reports the selected route in structured metadata. Speko Gateway is an open customer-side runtime that runs beside LiveKit or Pipecat, keeps BYOK provider credentials in the customer process, and exposes one local streaming protocol across providers. Gateway can use provider-direct routes, Speko-managed routes, or both, and its public repository owns the canonical wire contract consumed by the hosted Router. A separate hosted Agents API at api.speko.dev owns agents, sessions, phone calls, and telephony. Speko describes these surfaces as complementary but not interchangeable. Speko also publishes benchmark coverage by language, comparing speech-to-text and text-to-speech models across languages including English, Arabic, French, German, Hindi, Norwegian, Spanish, Tamil, and Telugu. The benchmark tables reference models from providers such as Deepgram, Soniox, OpenAI, Alibaba, Cartesia, Google, Smallest, Gradium, AssemblyAI, ElevenLabs, Gladia, Inworld, Modulate, xAI, and Meta, and score them against cost per stage and metrics such as word error rate and finalize latency. Commercially, Speko targets two buyer profiles. Router is for applications that want Speko to operate the managed STT, LLM, and TTS data plane. Gateway is for teams that own the conversation loop and want provider-direct streaming or local credential custody. The company routes product questions, support requests, security reports, billing questions, and partnership inquiries through team@speko.ai, and maintains a public support page and privacy policy covering its website, platform, APIs, SDKs, and demos.

• Speko is described as a Y Combinator company operated by Speko Group, Inc. from the United States.

• Speko Router is offered as a hosted, provider-neutral data plane at router.speko.dev with typed operations for model discovery, transcription, synthesis, and LLM responses.

• Speko Gateway is released as an open customer-side runtime that runs beside LiveKit or Pipecat and keeps BYOK credentials in the customer process.

• A separate hosted Agents API at api.speko.dev is offered for agents, sessions, phone calls, and telephony.

• Speko publishes language-by-language benchmark coverage for speech-to-text and text-to-speech models across nine languages.

Provider-neutral routing and gateway infrastructure for voice AI, emphasizing language-by-language benchmark coverage and the ability to route STT, LLM, and TTS calls by latency, quality, cost, or a balanced objective.

• Speko Router

  A hosted, provider-neutral STT, LLM and TTS data plane at router.speko.dev, with typed contracts and managed routing. Teams can let Router choose an eligible route by latency, quality, cost, or a balanced objective, or pin a provider and model; each response reports the selected route in structured metadata.

• Speko Gateway

  An open customer-side runtime that runs beside LiveKit or Pipecat, keeps BYOK provider credentials in the customer process, and exposes one local streaming protocol across providers. It can use provider-direct routes, Speko-managed routes, or both, and its public repository owns the canonical wire contract consumed by the hosted Router.

• Speko Agents API

  A separate hosted API at api.speko.dev that owns agents, sessions, phone calls, and telephony.

• Language-by-language speech model benchmarks

  Benchmark coverage by language for speech-to-text and text-to-speech, comparing models from providers such as Deepgram, Soniox, OpenAI, Alibaba, Cartesia, Google, Smallest, Gradium, AssemblyAI, ElevenLabs, Gladia, Inworld, Modulate, xAI, and Meta, scored against cost per stage and metrics such as word error rate and finalize latency.

1. 1The router for voice models | Spekospeko.ai
2. 2About Speko | Router and Gateway infrastructurespeko.ai
3. 3Spekowikidata.org