Projects

### Overview: VoxScribe is a free, open-source voice typing application designed for both Windows and Android devices. It allows users to dictate text directly into any app they are using, transcribing their speech locally without uploading to the cloud or requiring an account. The app is particularly useful for those who prefer to dictate without the need for an internet connection. ### Key Features: - **Windows:** Hotkey (F9 by default) for dictation anywhere, supports audio and video files conversion to text and subtitles. - **Android:** Floating button for dictation in any app, works in airplane mode. - **Offline Dictation:** Transcribes speech locally on the device, no cloud or account needed. - **Batch Import:** Automatically transcribes a folder of files, reports on progress, and handles skipped and failed files. - **Custom Vocabulary:** Allows adding custom vocabulary for names and jargon, and setting word replacements for common errors. - **Settings:** Offers options for text, voice, and language settings, including per-app profiles and language switching. ### Tech Stack: Python ### Why It Matters: VoxScribe is a valuable tool for anyone who needs to dictate text quickly and efficiently, particularly in situations where an internet connection is not available. The app's ability to transcribe speech locally and its user-friendly interface make it a practical solution for dictation needs on both Windows and Android devices. ### Repository: The project is hosted on GitHub at [ahmedhmam1994/voxscribe-ai-voice-dictation](https://github.com/ahmed

Overview: Linux Voice Assistant is a Python application designed to enhance productivity by providing voice commands for common tasks, including searching the web, managing applications, setting alarms, and more. It supports offline text-to-speech, making it useful even when internet connectivity is limited. Key features: - Voice commands for various tasks - Offline text-to-speech functionality - Integration with Google Translate for language translation - Ability to launch applications and update system Tech stack: Python Why it matters: This app simplifies daily tasks by allowing users to interact with their devices through voice commands, making it more accessible and convenient, especially for those with limited internet access or disabilities. Repository: aydinnyunus/LinuxVoiceAssistant Primary language: Python GitHub stars: 37

### Overview: This repository is a curated list of AI video generation APIs, SDKs, and tools. It covers a wide range of services and models for developers looking to integrate AI into their video applications. ### Key Features: - **Text-to-Video APIs**: APIs that convert text prompts into videos. - **Real-Time and Interactive Video**: Tools for creating live and interactive video experiences. - **Avatar and Talking Head APIs**: Platforms for generating video avatars and talking heads. - **Video Style Transfer and Motion**: Models for enhancing and simulating video styles and motion. - **Open Source Models**: Models that can be used without paying for access. - **SDKs and Developer Tooling**: Libraries and tools for developers to integrate AI into their projects. - **Infrastructure and Deployment**: Infrastructure for deploying and managing AI video generation services. - **Evaluation and Observability**: Metrics and tools for evaluating the quality and performance of generated videos. - **Templates and Example Projects**: Pre-built templates and example projects for developers to quickly start building video applications. ### Tech Stack: The repository uses Python, Node.js, and Go as the primary programming languages. ### Why it Matters: This list is essential for developers looking to integrate AI into their video applications, providing a comprehensive set of tools and models to create engaging and sophisticated video content.

### Overview: BuildArena 2.0 is a physics-grounded besiege building environment designed for LLMs. It provides an end-to-end engineering workflow, including automated game launch, machine construction, closed-loop control, telemetry, and timeline replay, all accessible via a Python interface and MCP tools. Powered by BuildArena ToolKit and Besiege CLI, agents can build, run, observe, and control machines in simulation. ### Key Features: - **End-to-End Engineering Workflow**: Automated game launch, machine construction, closed-loop control, telemetry, and timeline replay. - **Python Interface**: Accessible via a Python interface for easy integration and scripting. - **MCP Tools**: Utilizes MCP tools for enhanced control and simulation capabilities. - **Machine Construction**: Builds and runs machines in Besiege, observing and controlling them in simulation. ### Tech Stack: Python ### Why It Matters: BuildArena 2.0 offers a comprehensive solution for machine design and autonomous execution, leveraging Besiege's physics sandbox and Python tools. It streamlines the process from conceptual design to autonomous execution, making it ideal for AI agents in engineering and simulation contexts.

### Overview: OpenLive is an open-source, model-neutral runtime designed for real-time voice agents. It separates interaction continuity from model cognition, preserving native duplex capabilities and providing a robust, self-hosted environment. ### Key Features: - **Call-First Voice Surface:** A live call-first interface with a Bloub face on a black stage, auto-joining sessions, and settings for advanced features. - **Open AI Voice:** Production path to Piper TTS, auto engine selection, and licensing details. - **Client Audio Intelligence:** Noise suppression, VAD worklet, adaptive echo cancellation, and barge-in handling. - **Agent, Tools, & Sandbox:** Internal agent, sandbox file I/O, multi-agent pool, and durable user profile. - **Transport, Providers, & Tasks:** WebRTC fallback, coordinated WebRTC → WebSocket, Moshi duplex, and task lifecycle management. - **Desktop:** Tauri shells for Windows and macOS. ### Tech Stack: Rust ### Why It Matters: OpenLive offers a model-neutral, open-source solution for real-time voice agents, providing a self-hosted, deadline-sensitive environment that preserves native duplex capabilities and integrates seamlessly with various providers and agents. Repository: byte271/Openlive

Overview: ht is a shell command tool that answers questions about shell commands using AI. Key features: - Answers shell commands - Explains last answer - Explains commands - Executes last answer Tech stack: Dart Why it matters: Provides a simple way to understand and execute shell commands, making it useful for developers and system administrators. Repository: catallo/ht GitHub stars: 37 GitHub description: ht - a shell command that answers your questions about shell commands

### Overview: 本仓库提供了一套中文 9 段式人像提示词创作体系，适用于ComfyUI、Krea2和GPT Image 2等AI模型。该体系包含模板、风格预设、批量生成器和协调生成方法论，旨在稳定产出高质量的人像提示词。 ### Key features: 1. **模板与风格预设**：提供9段式模板，涵盖国风、现代、海外等多种风格预设，一键套用。 2. **批量生成器**：支持100+条提示词批量生成，包括服装、饰品、鞋履和背景的协调匹配。 3. **结构化提示词**：前4段为真实感内核，后5段按需求自由填充，便于复用。 4. **无强制约束**：创作自由，无强制校验，无自动合规拦截。 5. **跨平台适配**：模板适用于ComfyUI、Krea2和GPT Image 2等模型。 ### Tech stack: Python ### Why it matters: - 稳定产出高质量人像提示词，降低创作成本。 - 提供多样化的风格预设，满足不同需求。 - 支持批量生成，提高工作效率。 - 无强制校验，创作自由度高。 ### GitHub stars: 36

### Overview: Mimasa is a real-time multilingual face translator that uses advanced computer vision, machine learning, and natural language processing to provide seamless translation and synchronization of speech and facial expressions in real-time. ### Key Features: - Real-time translation of speech to another language. - Seamlessly tracks facial movements and speech in real-time. - Accurately separates music and speech from video input. - Provides a more natural and realistic translation experience. - Supports multilingual support. - Utilizes advanced computer vision and machine learning techniques. ### Tech Stack: Python ### Why It Matters: Mimasa aims to bridge the language barrier and make communication between people of different languages easier and more efficient. It provides a seamless translation experience that is both accurate and natural, making it a valuable tool for language learners, interpreters, and anyone working in international settings.

Overview: This project is an official implementation of the Conditional Brownian Bridge Diffusion Model for the translation of VHR SAR to Optical images. It builds upon the original work by Bo Li et al. and is based on the Latent Diffusion Model (LDM). Key features: - Conditional Brownian Bridge Diffusion Model for SAR to Optical Image Translation - Compatible with paired translation tasks - Preprocessing script for the MSAW dataset - Configuration templates for various datasets Tech stack: Python Why it matters: This project provides a practical solution for image translation tasks, specifically for SAR to Optical image conversion, which has applications in remote sensing and environmental monitoring. Repository: egshkim/ConditionalBBDM-for-VHR-SAR-to-Optical Primary language: Python GitHub stars: 37 GitHub description: The official implementation of Conditional Brownian Bridge Diffusion Model for VHR SAR to Optical Image Translation README excerpt: This is the official implementation of [Conditional Brownian Bridge Diffusion Model for VHR SAR to Optical Image Translation](https://arxiv.org/abs/2408.07947).

### Overview: nanoDD is a simple and scalable Discrete Diffusion implementation for text generation using PyTorch. This project aims to educate, research, and provide fun for beginners and experienced developers alike. ### Key Features: - Simple and efficient implementation of Discrete Diffusion for text generation. - Supports both training and sampling from pre-trained models. - Compatible with the text8 and openwebtext datasets. - Scalable training script for efficient model development. ### Tech Stack: PyTorch ### Why It Matters: - Educational tool for understanding Discrete Diffusion algorithms. - Simplified approach for beginners to work with discrete diffusion models. - Scalable training script for advanced users to experiment with larger models. ### Repository: flukeskywalker/nanoDD GitHub Stars: 37 GitHub Description: Simple Scalable Discrete Diffusion for text in PyTorch

### Overview: Upload personal docs and Chat with your PDF files with this GPT4-powered app. Built with LangChain, Pinecone Vector Database, deployed on Streamlit. ### Key features: - Upload PDF files - Chat with PDF content - Use LangChain for text processing - Utilize Pinecone for vector database - Streamlit for deployment ### Tech stack: Python ### Why it matters: This app leverages cutting-edge AI technologies to provide a personalized and interactive experience with PDF documents, enhancing productivity and knowledge retrieval.

Overview: This repository fine-tunes the Whisper model for Taiwanese speech recognition using the TAT dataset. Key features: - Fine-tuning Whisper model on the TAT dataset - Custom dataset preparation script for non-TAT datasets - Whisper tokenization without language token - Initial results after 4 epochs on TAT-Vol1-eval split Tech stack: Python Why it matters: This project aims to improve Taiwanese speech recognition by fine-tuning the Whisper model on a specific dataset, providing a preliminary result for further refinement.

### Overview: This repository contains the code for fine-tuning the DeepLabV3 model for multiclass semantic segmentation using PyTorch. The model is trained on a dataset of personal skydiving images, and the results are visualized on a test image. ### Key Features: - Fine-tuning of DeepLabV3 model for multiclass semantic segmentation - Training on a dataset of personal skydiving images - Results visualized on a test image ### Tech Stack: - Python ### Why it Matters: This project demonstrates the application of semantic segmentation techniques using the DeepLabV3 model, showcasing how it can be fine-tuned for specific use cases. The results provide a practical example of how to apply semantic segmentation to real-world data.

Overview: This project aims to generate a speech recognition dataset for Mandarin, leveraging the abundance of Taiwanese dramas on YouTube. The dataset is designed to address the lack of Mandarin speech recognition datasets, particularly those containing English or Taiwanese speech. Key features: - Automatically downloads videos from YouTube. - Uses FFMPEG to extract audio and split videos into frames. - Applies Mask-RCNN to identify subtitles in the frames. - Converts OCR results to SRT files for each video. - Generates a Jupyter notebook for Mask-RCNN training dataset. Tech stack: Python, TensorFlow, Keras, Google Cloud Vision, Pafy, YouTube-DL, tqdm, Editdistance. Why it matters: This project is crucial for researchers and developers looking to improve speech recognition models for Mandarin, as it provides a comprehensive dataset that can be used for training and evaluating models. Repository: khuangaf/ITRI-speech-recognition-dataset-generation Primary language: Jupyter Notebook GitHub stars: 37 GitHub description: Automatic Speech Recognition Dataset Generation

Overview: This project aims to automate the note-taking process during Microsoft Teams meetings, leveraging artificial intelligence to ensure all attendees remember the key points of the discussion. Key features: - AI-driven note-taking - Integration with Microsoft Teams - Automatic summarization of meeting content Tech stack: Python Why it matters: By automating the note-taking process, this project aims to reduce the cognitive load on team members, especially those working from home, and ensure that all meeting details are captured accurately and efficiently. Repository: kromme/Teams-Notetaker GitHub stars: 37 GitHub description: Let AI create the notes of your Teams Meeting

### Overview: This repository contains the open-source implementation of the cross attention mechanism from the paper "JOINTLY TRAINING LARGE AUTOREGRESSIVE MULTIMODAL MODELS". The code is designed for large autoregressive multimodal models and is compatible with Python. ### Key features: - Cross attention mechanism implementation - Supports multimodal inputs (e.g., text, images) - Open-source and compatible with Python ### Tech stack: Python ### Why it matters: This implementation is crucial for researchers and developers working on large-scale multimodal models, enabling efficient and effective processing of text and image data.

Overview: This is an end-to-end example showcasing how to use Kubeflow for both machine learning and model serving. Key features: - Building machine learning using Kubeflow's Jupyter - Collaborative filtering for product recommendations based on purchasing history - Exporting Python code for model serving using TFJob - Running two instances of TF-serving for model serving - Using Kubeflow pipelines to coordinate execution of steps Tech stack: Jupyter Notebook Why it matters: This example demonstrates the integration of Kubeflow with machine learning and model serving, providing a practical solution for implementing recommendation systems in a production environment.

Overview: WebVoice SDK is a lightweight and optimized library for building voice-enabled applications in the browser, providing hardware microphone handling, downsampling, speech preemphasis, features extraction, and wake word detection. Key features: - Hardware Microphone Handler - Downsampler - Speech Preemphaser - Voice Activity Detection - Features Extraction - Wake Word Detection Tech stack: JavaScript Why it matters: WebVoice SDK simplifies the development of voice-enabled applications by providing a robust and efficient solution for real-time voice processing, making it ideal for building applications that require voice input and interaction directly in the browser. Repository: linto-ai/WebVoiceSDK Primary language: JavaScript GitHub stars: 37 GitHub description: Buildings block for voice-enabled applications in the browser README excerpt: WebVoice SDK is a JavaScript library that provides lightweights and fairly well optimized buildings block for always-listening voice-enabled applications right in the browser. This library is the main technology behind LinTO's Web Client as it deals with everything related to user's voice input. The library comes prepacked with two wake word models and is available as ES module and IIFE bundle for direct browser usage. The wake word Engine relies on Tensorflow JS and WebAssembly portable runtime to infers towards single or multiple wake-words model with lightweight and ultra-effecient performances.

Overview: Aoororachain is a Ruby tool designed to work with Large Language Models (LLMs) and integrate them with LLM servers and private data. Key features: - Installation guide for Ruby and Open Source LLMs - Requisites for working with LLMs, including setup for Llama.cpp, LLM server, Open Source models, and Chroma DB - Python environment and Open Source embedding models installation instructions - Embedding model installation and usage instructions Tech stack: Ruby, Open Source LLMs, Llama.cpp, LLM server, Chroma DB, Python, LangChain, Sentence Transformers, Instructor Embedding Why it matters: Aoororachain simplifies the process of integrating LLMs with Ruby, making it accessible for developers working with private data and Open Source models.

### Overview: This project is an official implementation of the paper "Diverse Controllable Diffusion Policy with Signal Temporal Logic" presented at RA-L2024 and ICRA2025. The implementation leverages NuScenes dataset and Signal Temporal Logic (STL) for learning diverse and controllable policies in autonomous driving scenarios. ### Key features: - **Diverse Controllable Policy Learning**: The project aims to develop a policy learning framework that can generate diverse and controllable driving behaviors while satisfying STL specifications. - **NuScenes Data Visualization**: A NuScenes data visualizer and annotation tool are implemented, along with annotated data. - **Official Implementation**: The project provides an official implementation of the RA-L2024/ICRA2025 paper, focusing on diverse and controllable policy learning for autonomous driving. - **Data Collection and Preprocessing**: The project includes steps for data collection, trajectory optimization, and the training of the model from scratch. ### Tech stack: - **Python**: The primary programming language used for implementing the project. ### Why it matters: This project is significant because it provides a practical solution for learning diverse and controllable policies in autonomous driving scenarios, which is crucial for improving the safety and performance of autonomous vehicles. The use of Signal Temporal Logic (STL) ensures that the policies generated are not only diverse but also meet specific safety and performance requirements. ### Repository: The project is hosted on GitHub at [mengyuest/pSTL-diffusion-policy](https://github.com/mengyuest/p