Pioneering the next generation of Speech AI technology

Our research is focused on creating superhuman Speech AI models that will unlock entirely new classes of applications and products to be built leveraging voice data.

Wide banner image with a dark blue gradient background transitioning from deep to light blue from bottom to top. White text in the center reads 'Conformer-2'. Below the text, a visual representation of a soundwave in blue spans the width of the image, suggesting audio or speech patterns.Wide banner image with a dark blue gradient background transitioning from deep to light blue from bottom to top. White text in the center reads 'Conformer-2'. Below the text, a visual representation of a soundwave in blue spans the width of the image, suggesting audio or speech patterns.

Introducing Conformer-2

We're introducing Conformer-2, our latest AI model for automatic speech recognition. Conformer-2 is trained on 1.1M hours of English audio data, extending Conformer-1 to provide improvements on proper nouns, alphanumerics, and robustness to noise.

Read more
Wide banner with a dark background gradually brightening towards the center. On the left, there's an icon of a lemur face within a blue app square, next to the white text 'LeMUR'.Wide banner with a dark background gradually brightening towards the center. On the left, there's an icon of a lemur face within a blue app square, next to the white text 'LeMUR'.

Introducing LeMUR

Introducing LeMUR, the easiest way to build LLM apps on spoken data. Search, summarize, ask questions, and generate new text, with knowledge of all your application’s spoken data. LeMUR performs intelligent retrieval to offer high-quality LLM responses with a single API call.

Read more

All Research

  • Light-weight probing of unsupervised representations for Reinforcement Learning

    An evaluation protocol for unsupervised RL representations uses two linear probing tasks to predict rewards and expert actions, reducing computational cost and improving RL training efficiency.

    Read More
  • Introducing our new punctuation restoration and truecasing models

    We’ve trained new Punctuation and Truecasing models on 13 billion words to achieve a 39% F1 score improvement for mixed-case words. Building on a novel application of a hybrid architecture for a character-level classifier reduces inference time and improves the scalability of our Speech AI systems.

    Read More
  • Conformer-1: A robust speech recognition model trained on 650K hours of data

    We're introducing Conformer-1, a state-of-the-art speech recognition model trained on 650K hours of audio data that achieves near human-level performance and robustness across a variety of data.

    Read More
Careers

Leaders in Speech AI research

We believe the best way to continue to innovate is to bring together some of the best minds in AI across different fields, expertise, and backgrounds. Join our team of interdisciplinary research leaders, scientists, and engineers working to advance the state-of-the-art in AI models for voice data.