oruk

oruk is a research lab and developer platform focused on speech AI. According to its official site, the company trains speech foundation models designed to understand not only the words in an audio recording but also how those words are spoken — tone, rhythm, emphasis, emotion, intent, sarcasm, and speaking style. Rather than treating transcription as a purely text problem, oruk models return transcripts annotated with emotion and speaking-style labels, so that applications can respond to a hesitant answer, a frustrated customer, or an excited user. The company positions itself as a research lab first and a product platform second. It publishes evaluation methods, comparison results, and limitations so that researchers and developers can inspect the work before building on it. Its models are delivered through a single API, with SDKs, examples, and a playground for testing custom audio. The flagship offering highlighted on the site is Oruk Resonance 2, which is in preview with an early-access application process. Earlier work includes Resonance 1 and the Orukeet family of open speech models, including an Orukeet model covering 25 languages and a faster Orukeet variant. oruk targets developers and organizations that need speech understanding rather than plain transcription. Named use cases include voice agents, contact centers, sales teams, hardware and robotics, media and captions, meetings and research, games and experiences, and developer platforms. The pitch to builders is to "give your agent a better ear" — letting an agent pick up on hesitation, frustration, or excitement and use that context to shape its response. The company is backed by a16z speedrun, according to its About page. It is led by co-founders Nathan Roll (CEO) and Vianney Grenez (CTO), supported by research scientists, speech data specialists, and advisors. The lab also supports the oruk organization, a nonprofit arm that works to bring speech technology to endangered languages. oruk is a private company; no public listing or ticker is indicated in the available evidence.

oruk presents itself as a research lab for speech AI. Its stated mission is to train speech AI to understand words and how they are spoken, using tone, rhythm, and emphasis to return transcripts with emotion and speaking-style labels. The company frames this as a departure from conventional speech-to-text, which captures what was said but not how it sounded. The lab's public research output includes work on speech recognition models. An Orukeet model was introduced as an open speech model for 25 languages, using fitted Gabor filters, and a follow-up experiment described making Orukeet twice as fast while measuring the accuracy tradeoff. Another research piece, "A voice. Many meanings.," examined how small differences in speech acquire meaning within communities. These items were published on the company's site in September 2026. On the product side, oruk has moved from research models toward a developer-facing API. Resonance 1 was an earlier model, and Oruk Resonance 2 entered preview with an early-access application process. The company describes its models as available through one API, with SDKs, examples, and a playground for testing custom audio. It emphasizes published evaluations — evaluation methods, comparison results, and limitations — so that researchers and developers can inspect results before building. oruk is backed by a16z speedrun, according to its About page. The company is led by co-founders Nathan Roll (CEO) and Vianney Grenez (CTO), with a team that includes research scientists Irene Yi and Büşra Marşan, several speech data specialists, and advisors Calbert Graham, Jonas Hedberg, Corey Weiss, and Igor D'Angelo. The lab also supports the oruk organization, a nonprofit arm bringing speech technology to endangered languages. The company targets a broad set of applications: voice agents, contact centers, sales teams, hardware and robotics, media and captions, meetings and research, games and experiences, and developer platforms. Its core promise to builders is that agents can pick up on hesitation, frustration, or excitement and use that context to shape responses. No founding year, headquarters location, or public listing is stated in the available evidence.

• Resonance 1 is referenced as an earlier oruk model

• Orukeet introduced as an open speech model for 25 languages

• Experiment published on making Orukeet twice as fast

• Research piece "A voice. Many meanings." published

• Oruk Resonance 2 enters preview with early-access applications

A research lab for speech AI that models transcription, emotion, and speaking style together, with published evaluations, delivered to developers through one API.

• oruk speech understanding API

  A single API delivering speech models with SDKs, examples, and a playground.

• Oruk Resonance 2

  Flagship model in preview, available via early access.

• Resonance 1

  Earlier oruk model referenced on the site.

• Orukeet

  Open speech model for 25 languages using fitted Gabor filters, with a faster variant.

1. 1oruk — Speech API for transcripts, emotion, and styleoruk.ai
2. 2About oruk labs | orukoruk.ai
3. 3orukwikidata.org