Karen Livescu
Papers
3
Total Citations
64
H-Index
2
About
Karen Livescu is a leading researcher in speech and language processing, with a focus on visually grounded speech models and low-resource speech technology. Her major contributions lie in developing methods that allow machines to learn from untranscribed speech paired with visual context, such as images, mirroring how infants acquire language. This work is pivotal for applications in robotics, human language acquisition, and low-resource speech processing. Her most cited paper, "Semantic speech retrieval with a visually grounded model of untranscribed speech" (2017, 50 citations), demonstrates how models can map speech and images into a shared semantic space, enabling tasks like image retrieval from spoken queries without textual transcripts. She has also advanced keyword prediction from untranscribed speech, as seen in her related works (2017, 12 and 2 citations), which explore learning meaningful representations from audio-visual data. Livescu’s research bridges the gap between speech recognition and multimodal learning, offering scalable solutions for under-resourced languages and settings. Her innovative approach to grounding speech in visual perception has made her a key figure in the emerging field of visually grounded language learning.
Research Focus
Key Achievements
Top Papers
- 1
- 2Visually Grounded Learning of Keyword Prediction from Untranscribed Speech12 citations · 2017
- 3Semantic keyword spotting by learning from images and speech.2 citations · 2017