According to a recent LinkedIn post from LiveKit, the company is showcasing AssemblyAI’s Universal-3 Pro speech-to-text model integrated via LiveKit Inference and tested through a custom LiveKit Agent. The post outlines four streaming test scenarios focused on entity accuracy, multilingual code-switching, domain-specific vocabulary, and verbatim disfluency capture.
Claim 55% Off TipRanks
- Unlock hedge fund-level data and powerful investing tools for smarter, sharper decisions
- Discover top-performing stock ideas and upgrade to a portfolio of market leaders with Smart Investor Picks
The LinkedIn post highlights that Universal-3 Pro appears to improve handling of structured data such as credit card numbers, emails, and phone numbers versus a prior Universal Streaming model, and that language detection can support mixed English-Spanish input without forcing a single language in transcripts. The model is also described as benefiting from keyterms prompts for specialized terminology, with AssemblyAI reportedly citing up to 45% accuracy gains for targeted terms.
As shared in the post, the tests were monitored using LiveKit Cloud’s observability dashboard, enabling side-by-side transcript comparison rather than relying on subjective evaluation. Universal-3 Pro is presented as accessible directly through LiveKit Inference using an existing LiveKit API key, suggesting a streamlined adoption path for developers already on the LiveKit platform.
For investors, the post suggests LiveKit is deepening its positioning in real-time AI communications by integrating higher-accuracy, promptable transcription within its infrastructure stack. If customers adopt these capabilities for use cases in legal, medical, contact centers, or coaching, this could support higher usage of LiveKit Cloud, improve developer stickiness, and potentially expand the company’s monetization opportunities in AI-powered voice and video applications.