Vocal characteristics play a significant role in constructing and perceiving individual identity. Conditions such as amyotrophic lateral sclerosis, Parkinson's, and multiple sclerosis can cause a degradation of unique voice characteristics over time. Additionally, conditions like muscular dystrophy and profound deafness can limit an individual's ability to produce certain sounds or impact vocal and articulatory patterns.

Source: https://blog.research.google/

Large language models struggle with factual inaccuracies and hallucinated content, particularly with knowledge-intensive questions. Retrieval augmented generation (RAG) is a promising solution that leverages an external knowledge base to retrieve documents with related information and incorporates it into generated responses.

Source: https://blog.research.google/

YouTube Music boasts a catalog of over 100M songs globally, and its recommender system uses user actions such as skip, like, or dislike to understand a user's musical taste. This signal is critical to understanding a user's preferences and providing better recommendations.

Source: https://blog.research.google/

Komal Singh, a researcher, grew up in India where she was influenced by the media she consumed, which often lacked representation of people who looked like her. She partnered with the Geena Davis Institute, IAA India Chapter, and University of Southern California to apply AI models that understand representation in the media. The goal is to create a future where people globally see themselves reflected in the stories they watch and read.

Source: https://blog.research.google/

Source: https://blog.research.google/

Recent advancements in Large Language Models (LLMs) have led to the development of multimodal LLMs like Gemini, which can process images and videos while retaining language understanding and generation capabilities. However, the issue of object hallucination, where generated language includes descriptions of objects not present in the input, presents a significant challenge to their adoption. For example, LLaVA-v1.5 inaccurately mentioned the presence of a toothpick in an image.

Source: https://blog.research.google/

We present a method that allows for parametric editing of material properties, such as color, shininess, or transparency, of objects in any photograph. This method leverages the real-world understanding of generative text-to-image models through fine-tuning with a synthetic dataset. The resulting parametric model provides precise control over how shiny an object appears or the exact shade of paint on a wall.

Source: https://blog.research.google/

We describe an approach for using photoplethysmograph (PPG) data for potential use in early detection of cardiovascular disease risk. Heart attack, stroke, and other cardiovascular diseases remain the leading cause of death worldwide, claiming millions of lives each year. The approach aims to make heart health screening accessible by releasing tooling to help people collect PPG signals with their smartphone.

Source: https://blog.research.google/

Scientists have developed NeuralGCM, a machine learning-based approach to simulating Earth's atmosphere, in partnership with the European Centre for Medium-Range Weather Forecasts (ECMWF). This new method combines traditional physics-based modeling with machine learning to make accurate predictions about the climate. The goal is to answer questions about the future, such as where droughts and tropical storms will occur, and how wildfire seasons will change.

Source: https://blog.research.google/

Google has employed special infrastructure for large-scale code migrations, using static analysis and tools like Kythe and Code Search to discover which locations need to be changed along with their dependencies. The infrastructure also employs tools like ClangMR to make the changes. This approach works well for changes that are uniform in their structure and have a limited set of edits.

Source: https://blog.research.google/