Toucan TTS, developed by the Institute for Natural Language Processing (IMS) at the University of Stuttgart, supports speech synthesis in more than 7,000 languages. This advanced text-to-speech toolbox is designed to transform the field of multilingual TTS systems, making it an invaluable resource for educators, researchers, and developers3.
Luna is a lightweight DeBERTA-large encoder fine-tuned for hallucination detection in Retriever-Augmented Generation (RAG) settings6. It evaluates the entailment of each context sentence with the generated response, labeling tokens in unsupported sentences as hallucinations. Luna outperforms zero-shot prompting and RAG evaluation frameworks, offering competitive performance on RAGTruth benchmark while significantly reducing cost and latency.
The GenQA Instruction Dataset, created by researchers from the University of Maryland, aims to automate the generation of large-scale instruction datasets for AI model finetuning and diversity enhancement. By using a single prompt, the method generates diverse sets of instruction examples with minimal human oversight, benefiting industrial-scale datasets and enhancing AI research and applications4.