Company profile · 2 min read
Oruk: speech analysis for transcripts, emotion and speaking style
Oruk offers speech models and an API for transcription, emotion and speaking-style analysis.
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
Published · Updated
Oruk provides speech analysis APIs for developers building voice analytics, QA and speech-aware products.
What it does
The public documentation describes realtime speech in 32 locales, prerecorded English analysis, phrase-level emotion and style labels. Pricing lists Hobby at $9/month, Builder at $49/month and Production at $199/month, with included minutes and a seven-day trial.
Buyer and task
Developers building voice analytics, QA or speech-aware products that need more than transcription.
Workflow boundaries
The checked public pages do not independently measure accuracy, latency or emotion validity. Plans and supported locales are page claims checked in the September 20, 2026 page snapshot.
What I would ask
Ask which analyses are production-ready, how audio is retained, and how a customer should interpret an emotion or style label.
Why it fits
Oruk is a developer-facing speech layer that goes beyond transcription. The homepage names speech-to-text, emotion, intent, tone, sarcasm and understanding. The docs separate prerecorded English analysis from realtime speech, with 32 supported locales for live transcription and phrase-level emotion scores. The contact-center page describes a practical integration: send a completed call, receive transcript segments plus emotion and speaking-style scores, link passages back to playback and queue candidates for human review.
The pricing page makes the initial test unusually clear. Hobby is $9/month, Builder $49/month and Production $199/month, each with different speech-understanding and Orukeet transcription allowances, API-key counts and support. The docs also describe production limits and request IDs. Those are useful implementation details, but they do not establish that an emotion label is valid for a particular accent, recording condition or business decision.
A contact-center team should begin with a labeled review sample and keep model, request and recording IDs with every result. Compare useful passages found and false positives against the current sampling process before turning a score into a coaching or customer-impact decision. The public pages describe the API, but do not establish a measured accuracy result across accents, recording conditions or business decisions.
Quick facts
| Field | Sourced detail |
|---|---|
| Buyer | Developers building voice analytics, QA and speech-aware products |
| Outputs | Transcript, emotion, intent and speaking-style signals |
| Localized realtime | 32 supported locales named in the docs |
| Plans | Hobby $9/mo; Builder $49/mo; Production $199/mo, with separate allowances |
Sources checked
official Speedrun profile · company homepage · source page · source page · source page
Sources checked — September 20, 2026.
About the author
Cofounder of Lazyweb (a16z Speedrun). Previously grew subscriptions and built native ads at Duolingo. Writes First1000 (90,000+ subscribers).
I cofound Lazyweb and publish Mudpie. This is an owner-written publication, not an independent testing organization. Research notes distinguish observations, sourced reporting and editorial judgment.
First1000 ↗ · X ↗