Assembly AI claims its new Universal-1 model has 30% fewer hallucinations than Whisper

Key Points:

  • Assembly AI introduces Universal-1 speech recognition model trained on 12.5 million hours of multilingual audio data.
  • Universal-1 provides accurate transcription across multiple languages and improved timestamp estimation.
  • Universal-1 shows increased efficiency in parallel inference, processing data five times faster than its competitors.


Assembly AI, an AI-as-a-service provider, introduces its new speech recognition model, Universal-1. Trained on vast amounts of multilingual audio data, Universal-1 excels in speech-to-text accuracy across four major languages and reduces errors in speech data and ambient noise significantly compared to existing models. The model, capable of transcribing multiple languages in a single audio file, also offers enhanced timestamp estimation, improved speaker diarization, and faster processing of long audio files.


The Universal-1 model, proclaimed as a milestone in providing accurate speech-to-text capabilities, is highlighted for its superior benchmark performance compared to industry peers. Assembly AI emphasizes the practical applications of Universal-1, such as generating precise notes, identifying action items, sorting metadata, and enhancing various AI-powered applications like video editing workflows and telehealth platforms.


Moreover, Assembly AI is organizing the AI Impact Tour in Atlanta on April 10th, focusing on how generative AI is revolutionizing the security workforce. This exclusive event, in collaboration with Microsoft, offers invitees the opportunity to explore the innovative uses of AI in security and network with industry experts. Interested individuals can request an invite to the event to gain valuable insights into the transformative potential of AI in security operations.


For those interested in leveraging the benefits of improved speech-to-text AI models like Universal-1, Assembly AI provides access through its API. By incorporating accurate and efficient speech recognition technology, organizations can enhance productivity and accuracy in various applications, from note-taking to clinical documentation and claims processing.
Get the latest news and updates: Subscribe to VentureBeat’s newsletters to stay informed about the latest advancements in AI, tech, and business.



Prompt Engineering Guides



©2024 The Horizon