🚀 What is Trending in AI Research?: PromptTTS 2 + PhysObjects + CoALA + BigVSAN + Verba + What is Trending in AI Tools? .....

Hey Folks!

This newsletter will discuss some cool AI research papers and AI tools. But before we start, we have included a small message from our sponsor.

➡️ Microsoft Researchers Unveil PromptTTS 2: Revolutionizing Text-to-Speech with Enhanced Voice Variability and Cost-Effective Prompt Generation

This paper from Microsoft introduces PromptTTS 2, which aims to tackle two major issues: the inability to fully describe voice variability through text prompts (the one-to-many problem) and the limited availability of text prompt datasets. PromptTTS 2 employs a "variation network" that predicts voice attributes not captured by text prompts. It also features a prompt generation pipeline that uses a speech understanding model and a large language model to formulate high-quality text prompts for speech. Testing on a 44K-hour dataset shows that PromptTTS 2 outperforms previous methods in generating voices consistent with text prompts and allows for diverse voice sampling, thus providing users with more voice-generation options. Importantly, the prompt generation pipeline reduces the need for costly manual labeling.

➡️  Meet PhysObjects: An Object-Centric Dataset of 36.9K Crowd-Sourced and 417K Automated Physical Concept Annotations of Common Household Objects

This paper from Stanford introduces PhysObjects, a comprehensive dataset containing both crowd-sourced and automated annotations related to the physical attributes of common household items. By fine-tuning a VLM on this dataset, the authors show that the model gains a more nuanced understanding of object properties like material and fragility. The enhanced VLM is then integrated into an interactive framework alongside a language model-based robotic planner. The results indicate significantly improved performance in planning tasks that require understanding physical object concepts, outperforming baseline models. The approach even demonstrated better task success rates in real-world robotic tests.

➡️  Princeton Researchers Propose CoALA: A Conceptual AI Framework to Systematically Understand and Build Language Agents

What is the best way to systematically design language agents that can perform tasks requiring grounding or reasoning? This paper from Princeton University presents a conceptual framework called Cognitive Architectures for Language Agents (CoALA). Drawing from the rich history of agent design in symbolic AI, it aims to systematize the development of large language models (LLMs) that can interact with external resources or use internal control flows like prompt chaining. The authors argue that LLMs share many properties with production systems, a class of symbolic AI systems. CoALA serves as a blueprint to bring together diverse methods for reasoning, grounding, learning, and decision-making in LLMs. The framework also highlights existing gaps and proposes future research directions for creating more capable language agents.

➡️ Researchers from Sony Propose BigVSAN: Revolutionizing Audio Quality with Slicing Adversarial Networks in GAN-Based Vocoders

How can the performance of Generative Adversarial Network (GAN)-based vocoders be improved for synthesizing high-fidelity audio waveforms? This paper from Sony investigates the effectiveness of Slicing Adversarial Network (SAN), a modified GAN framework, in enhancing vocoding tasks. The authors propose a modification to the Least-Squares GAN, which is commonly used in vocoders, so as to make it compatible with the SAN framework. This involves altering the loss functions to meet SAN's requirements. Experimental results indicate that this SAN-adapted approach can enhance the performance of existing GAN-based vocoders, including BigVGAN, with only minor adjustments to the architecture.

➡️ Meet Verba: An Open-Source Tool to Build Your Own RAG Retrieval Augmented Generation Pipeline and Utilize LLMs for Internal-Based Outputs

Verba is an open-source project to provide RAG apps with a simplified, user-friendly interface. One can dive into the data and start having relevant conversations quickly. Verba is more of a companion than a mere tool regarding data querying and manipulation. Paperwork, comparison, and contrast between several sets of numbers, and data analysis- through Weaviate and Large Language Models (LLMs), Verba enables all of this to be achievable. Based on Weaviate’s cutting-edge Generative Search engine, Verba automatically pulls the necessary background information from the documents whenever a search is performed. It uses the processing power of LLMs to provide exhaustive, context-aware solutions.

What is Trending in AI Tools?

Editor’s Recommended AI Tool

Height 2.0 — the autonomous project collaboration tool powered by AI. Automate the repetitive tasks and processes you spend time on every week, like hosting standups and crafting release notes. Carve out more time in your week for the projects that inspire you most (and move the needle on your goals). Effortlessly keep your list and workspace organized with AI built right into the projects and tasks you’re working on — so instead of creating more work for you, project management frees you to work toward your goals.[Productivity]