Introducing Universal-2: The latest advancement in Speech-to-Text technology. Capture the complexity of human speech, enhanced transcript quality, and better conversational insights by tapping into the next generation of Speech AI.
Five months ago, we introduced Universal-1, our best-in-class Speech-to-Text model that achieved remarkable robustness and accuracy across several industry-critical benchmarks. Today, we're launching Universal-2, which builds upon that foundation to address critical "last-mile" challenges in conversational data.
Universal-2 pushes the boundaries of Speech-to-Text technology by improving accuracy in areas that matter most for real-world applications:
⭐️ 24% improvement in the recognition of names, brands, locations, and more
⭐️ 21% increase in accuracy across critical data like phone numbers, zip codes, and other numerical identifiers
⭐️ 15% improvement in transcript structure with proper punctuation and casing for things like emails, dates, and dollar amounts
And most importantly, nearly 3/4s of users prefer Universal-2 over other Speech AI models.
Universal-2 bridges the gap between raw accuracy metrics and what makes transcripts truly useful—precise formatting, accurate proper nouns, and natural-looking output that businesses can trust.
Try it out and let us know what you're building with it!
Congratulations on the launch, @nick_morris4 and Team! Very cool name as well😉. STT model will only gain importance, and quality might be the main differentiator. Keep up the great work!
The accuracy boost on proper nouns and numbers is a game-changer! 🎯 Been using Universal-1 for our customer call transcripts and those tricky company names and phone numbers were always a pain point. Just tested Universal-2 with some tough audio samples and wow - the difference in formatting and accuracy is night and day. Super impressed with how it handles those "last-mile" details that actually matter in real-world use. Worth the upgrade! 💪
Is it possible to read my knowledge base and extract some professional terminology keyword lists so that common terms in my industry or company can be targeted in subsequent speech material recognition?
Hey @an_zuo! Universal-2 is trained on millions of hours of diverse industry audio, so it’s ready to handle specialized terminology out of the box. We recommend giving it a try and seeing if it hits the mark without any customization. For added precision, our features like Custom Spelling can further boost accuracy with unique terms.
@linda_yanlin Thank you so much for your kind words! 😊 It means a lot to hear that you're finding value for both meeting minutes and academic work. We specifically designed it with versatility in mind, so it's wonderful to see it helping across different use cases. Your support on our launch day is truly appreciated!