Google Research
·
4 months ago
Google Research released WAXAL, an open-access speech dataset covering 27 Sub-Saharan African languages spoken by over 100 million people. The dataset includes approximately 1,846 hours of transcribed speech for automatic speech recognition and over 565 hours of high-fidelity audio for text-to-speech synthesis, developed in collaboration with African academic institutions. The resource aims to enable development of voice-enabled technologies in African languages that have previously lacked sufficient training data.
Google Research
·
4 months ago
Google released SpeciesNet, an open-source AI model that identifies nearly 2,500 animal species in camera trap images, trained on 65 million labeled images. The model achieves 99.4% accuracy in detecting animals and 83% accuracy in species classification, processing up to 250,000 images daily on a gaming GPU. Research groups worldwide now use SpeciesNet to analyze millions of wildlife images for conservation, with projects in Colombia, Australia, Idaho, and Tanzania reducing manual identification work from decades to days.
OpenAI Blog
·
4 months ago
Codex Security, an AI application security agent, entered research preview to detect and patch software vulnerabilities by analyzing project context. The tool aims to reduce false positives compared to traditional static analysis security testing by understanding code relationships and dependencies. Organizations can now test whether the agent improves vulnerability detection accuracy while reducing alert noise from existing security scanners.
OpenAI Blog
·
4 months ago
Balyasny Asset Management built an AI research system that combines model evaluation, OpenAI integration, and agent workflows to conduct investment analysis. The system processes information across OpenAI's full platform capabilities to generate research insights at scale. This approach reduces manual research effort and allows analysts to focus on higher-level decision-making rather than data gathering.
OpenAI Blog
·
4 months ago
Descript built a system using OpenAI's reasoning models to automatically dub videos into multiple languages while preserving original timing and meaning. The company processed large content libraries through this approach, enabling localization at scale without manual adjustment of speech synchronization. Video creators can now reach international audiences without separately re-recording dialogue for each language version.
