In the last few years we've seen an explosion of audio data available online. This coupled with advances in AI technology have allowed organizations to unlock the value of voice data in ways that were previously impossible. As a result, we've seen organizations build new products, services, and capabilities that serve millions of people around the world. Today, we’re announcing Universal-1, our most powerful and accurate model to date, trained on 12.5M hours of multilingual audio data to help power the next generation of Speech AI products and features. Some key stats on Universal-1: • 72% preferred to our most recent model Conformer-2 in human evals • 71% better speaker count estimation and 14% better word timestamp estimation compared to our prior models • Up to 30% fewer hallucinations than seq2seq models like Whisper • Just 38 seconds to process 1 hour of audio Learn more about Universal-1 on our blog: https://lnkd.in/e5inQ-x9
Congrats! 🚀 🚀
"Impressive advancements! Excited to see how Universal-1 will elevate Speech AI capabilities across diverse applications."
GPT-4o/multimodal dataset collection pipeline & service by Olewave.com; large-scale, from different languages and topics, 1/10 in data pricing with 5x data efficacy vs. other vendors. 20 yoe in speech/voice/audio R&D.
4moCongrats, AssemblyAI! Olewave delivers customized, labeled, and validated large-scale real-world NLP/CV/speech/multimodal datasets of various scenarios such as dictation and conversation in multi accents/dialects/languages, and of diverse topics such as education, finance, legal, entertainment, healthcare, retail, and customer service.