Uploaded October 2024 | Updated September 2026, 2 weeks ago

---

Universal: A next-gen speech-to-text model pushing beyond traditional WER (word error rate) metrics. Built on Universal-1's industry-leading performance in just 6 months.

Key results:
24% better at recognizing proper nouns
21% improvement in alphanumeric accuracy
15% enhanced text formatting
73% of users prefer Universal-2 compared to Universal-1
Overall more accurate and robust model especially on real-world speech complexity
Sets new standards across human and technical benchmarks

Architecture:
Smart architecture choices prioritized over simply scaling model size
Universal-2 uses a 660M parameter Conformer RNN-T model
Built an innovative all-neural formatting pipeline
Solved critical challenges like repeated token handling in RNN-T

Announcement Landing Page: assemblyai.com/universal-2
Try it yourself: assemblyai.com/playground
Google colab: colab.research.google.com/drive/1IP_RFufO_-iQVICDEtTbqqHSTgqWPNmD?usp=sharing

▬▬▬▬▬▬▬▬▬▬▬▬ CONNECT ▬▬▬▬▬▬▬▬▬▬▬▬

🖥️ Website: assemblyai.com
🐦 Twitter: twitter.com/AssemblyAI
👽 Reddit: reddit.com/r/assemblyai
▶️ Subscribe: youtube.com/c/AssemblyAI?sub_confirmation=1
🔥 We're hiring! Check our open roles: assemblyai.com/careers

▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬

#MachineLearning #DeepLearning