A framework for building end-to-end automatic speech recognition systems, providing reference implementations of 20+ ASR models and recipes for multiple languages.

A PyTorch-based automatic speech recognition framework that supports streaming and non-streaming inference across multiple model architectures and languages.

React Native bindings for on‑device Whisper (and NVIDIA Parakeet) speech‑to‑text, with GPU/Core ML acceleration, voice‑activity detection, and realtime streaming support.

A high-performance speech-to-text solution based on fine-tuned Whisper models, optimized for low-latency streaming and on-device inference on NVIDIA GPUs and Apple Silicon.

A TensorFlow-based framework for Automatic Speech Recognition that implements various ASR architectures and supports TFLite conversion for efficient deployment.

A PyTorch implementation of the Conformer architecture that combines CNNs and Transformers to improve speech recognition by capturing both local and global audio dependencies.

A voice chat application powered by the Moshi speech-to-speech model, providing low-latency, bidirectional audio streaming for human-like interaction.

wav2letter++ is an end-to-end automatic speech recognition framework that provides recipes and pre-trained models to implement state-of-the-art speech-to-text architectures.

A Python library that provides a unified interface for speech recognition, supporting multiple online and offline engines like OpenAI Whisper, Google Speech, and Vosk.

The official Node.js SDK for ElevenLabs, enabling developers to integrate lifelike AI voice synthesis, real-time audio streaming, and voice-powered AI agents into their applications.

A ComfyUI custom node integration for VoxCPM, a tokenizer-free TTS system that enables expressive speech generation, natural language voice design, and advanced voice cloning.

A unified API for open-source and cloud-based Speech-To-Text and Text-To-Speech services, simplifying audio processing for chatbots and voice applications.

AI Stock Master is a Windows/macOS desktop app that blends quantitative stock‑analysis algorithms (RTSI, TMA, MSCI) with a locally‑run LLM (Mini Ollama) to generate natural‑language market reports and trading signals for Chinese, Hong‑Kong and US equities. It’s a research‑oriented tool, not a commercial trading platform.

MemoMind is a locally hosted, GPU‑accelerated memory system for AI coding agents. It stores extracted facts from chats, documents, and daily life events in PostgreSQL + pgvector, builds a knowledge graph, and offers fast 4‑way hybrid retrieval (semantic, BM25, graph, temporal). The agent can retain new information, recall relevant memories, and reflect across the whole store. All data stays on the user’s machine, works with any OpenAI‑compatible LLM, and includes a web dashboard for browsing and exporting memories.

DynaSaur is an AI agent framework that writes and reuses Python code as actions to solve tasks, adapting dynamically when predefined tools are insufficient.

Paritok is an open‑source proxy that sits between coding agents (Claude Code, Cursor, Codex, OpenHands, etc.) and LLM APIs. It reduces input‑token usage by (1) filtering irrelevant tool schemas, (2) compressing file reads/tool outputs/history with a 4 B model, and (3) summarising old turns. The gateway is a Python package (`paritok[proxy]`) that can run locally (via Ollama or vLLM) or via a hosted GPU service. Reported savings are ~25 % on the first turn and >60 % in longer sessions, translating to noticeable cost reductions. The project is Apache‑2.0 licensed and includes a dashboard and VS Code extension.

A minimalist framework for benchmarking speech-to-text engines across various datasets to compare accuracy, latency, and computational efficiency.

AttackGen is an open‑source, Streamlit‑based web app that uses LLMs (via LiteLLM) together with MITRE ATT&CK, MITRE ATLAS and an AI‑insider‑threat model to generate fully‑fledged incident‑response tabletop scenarios, detection‑and‑response reports, and ATT&CK Navigator layers. It supports multiple LLM providers, offers a chat‑assistant for refinements, can be run locally or as a Docker container, and includes an optional FastMCP server for agentic integration.

AgentSphere is a Java‑Spring/React platform that runs LLM‑driven ReAct loops, supports multi‑agent sub‑runs, real browser automation via a Chrome extension, model routing, persistent multi‑level memory, human‑in‑the‑loop clarification, OIDC SSO, and an embeddable chat widget.

A Flutter plugin that enables apps to access native device speech recognition for short voice commands and phrases across Android, iOS, and web.

Leopard is an on-device speech-to-text engine that provides private, accurate, and computationally efficient transcription across multiple platforms.

An on-device streaming speech-to-text engine that provides private, real-time audio transcription across multiple platforms.

A self-hosted, high-performance speech-to-text API powered by whisper.cpp that is Deepgram-compatible for easy integration.

The official Python SDK for Deepgram, providing easy integration of automated speech recognition, text-to-speech, and language understanding APIs.

memory‑lancedb‑pro is a production‑grade OpenClaw plugin that gives AI agents persistent, searchable memory. It auto‑captures conversation snippets, classifies them with an LLM, stores them in a LanceDB vector+BM25 index, and automatically injects the most relevant memories into prompts. Features include hybrid retrieval, cross‑encoder reranking, Weibull‑based forgetting, per‑agent/user/project scopes, and support for any OpenAI‑compatible embedding provider. Install via a one‑click script or the OpenClaw CLI, configure with a short JSON block, and manage the store with a built‑in CLI.

Neighbor is a Rails gem that adds vector‑based nearest‑neighbor search to ActiveRecord models. It supports PostgreSQL (pgvector or cube), MariaDB, MySQL (HeatWave), SQLite, Redis, and S3, offering many distance metrics, multiple vector types, and both exact and approximate indexing. Use it to build semantic search, recommendations, or hybrid retrieval directly inside a Rails app.

A collection of PyTorch scripts and tools for speech processing, focusing on neural vocoders and speech spoofing countermeasures.

Elyra is a collection of JupyterLab extensions that add AI‑focused features such as a visual pipeline editor, batch‑job execution for notebooks/Python/R scripts, reusable code snippets, LLM‑powered code assistance, hybrid runtime support via Jupyter Enterprise Gateway, and Git integration. Installable via pip or conda (or via Docker), it lets data‑science teams build, run, and version‑control AI pipelines without leaving the notebook environment.

PaperForge Research OS v3 is a Python‑based tool that unifies AI‑assisted paper writing, experiment orchestration, artifact tracking, and verifiable publishing. It stores every claim‑to‑evidence link in a SQLite “Scientific Memory”, enforces three execution profiles (writing‑only, research, full), and supports local, Docker, SSH, Slurm, Kubernetes and cloud compute back‑ends. The CLI (`paperforge …`) and a local web UI let users run workflows, approve proposals, compile LaTeX with built‑in templates (generic, CVPR, IEEE, Elsevier), and release a deterministic source bundle after secret‑scanning. Designed for reproducible academic work, it is open‑source under a non‑commercial license.

A command-line TTS tool that provides a macOS-style 'say' experience using ElevenLabs or 60db for high-quality AI speech synthesis.

LightRFT is an open‑source, distributed reinforcement‑learning framework for fine‑tuning large language and multimodal generative models. It unifies training back‑ends (DeepSpeed, FSDP) with rollout engines (SGLang, vLLM), supports a range of RL algorithms (GRPO, REINFORCE++, DAPO, etc.), and works across text, image, video, and audio modalities. The library provides ready‑to‑run examples (e.g., GSM8K with Qwen2.5), extensive docs, and Docker images, making it suitable for researchers and engineers building RL‑based alignment pipelines.

A high-performance Rust-based server that provides a unified API for real-time Speech-to-Text and Text-to-Speech services across multiple providers.

ojAlgo is a pure‑Java library for high‑performance linear algebra, numerical optimisation (LP/QP/MIP), and lightweight data‑science tools (neural nets, clustering). Zero external dependencies, MIT‑licensed, and suitable for Java‑based AI/ML workloads.

A voice-to-text utility for Apple Silicon Macs that transcribes speech on-device and pastes the result into any active application.

A Python library and command-line tool for converting text into phonetic transcriptions across multiple languages using various backends.

cuPyNumeric is an NVIDIA library that implements the NumPy API on the Legate runtime, enabling NumPy code to run on GPUs and distributed clusters. It offers drop‑in compatibility, Linux Python 3.11‑3.14 support, and can be installed via Conda or PyPI. The project is now end‑of‑life (final version v26.06.01) but remains available for historical use under the Apache‑2.0 license.

Flock is a Docker‑based service that runs a Claude‑Code powered dev team you control via chat (Telegram, VK, or a text‑only adapter). From a single message it plans, writes, tests, reviews and opens a PR, with optional loops for self‑verification, goal evaluation, scheduled jobs and CI watching. Each chat gets its own isolated workspace and branch, and the system works with GitHub/GitLab/Gitea using a PAT. Deployment is a one‑line `docker compose up -d` after filling a small `.env` file; the core logic lives in `core/agents/` and adapters are thin wrappers.

A local-first macOS app for Apple Silicon that provides private, on-device transcription for meetings, voice memos, and system-wide dictation, outputting structured Markdown files.

QuickVoice is an open‑source, self‑hostable platform for building AI voice agents (inbound/outbound calls) with full control over the stack—marketing site, console, Express API, Python LiveKit worker, telephony (Twilio/Telnyx), RAG (Pinecone), billing (Stripe), and privacy features. It runs locally via a single `task up:dev` command and is designed for teams that need inspectable, extensible phone‑automation infrastructure.

A native macOS Swift app that connects directly to Amazon Bedrock, letting users chat with text, image and other foundation models, attach files, run local tools/skills, and schedule automated prompts—all with data stored locally on the Mac.

QodeAssist is a GPL‑v3 Qt Creator plugin that adds AI‑driven code completion, chat, and inline refactoring for C++/QML. It works with local (Ollama, llama.cpp, LM Studio) and cloud LLM providers (Claude, OpenAI, Gemini, Mistral, Qwen, DeepSeek, etc.) and exposes its project‑aware tools via the Model Context Protocol (MCP) for use by other editors. The project is archived and no longer maintained.

Mulch is a CLI‑driven, git‑backed knowledge base that lets AI agents record structured learnings (conventions, failures, decisions, etc.) in per‑domain JSONL files and later retrieve context‑optimized excerpts for prompt injection. It supports custom record types, evidence‑based scoping, health‑checks, and pruning, enabling teams to accumulate and share expertise across agent sessions.

A local-first voice AI platform providing a desktop app, CLI, and server for speech-to-text, text-to-speech, and chat workflows without requiring cloud APIs.

A streamlined deployment wrapper for ChatTTS that provides a WebUI and one-click Colab setup for advanced text-to-speech generation.

Klaat Code is a terminal‑based AI coding assistant that talks to the hosted Klaatu‑o1 router. The router automatically selects the cheapest model tier for each request and escalates only when needed, cutting costs dramatically. The client indexes your project into a semantic call‑graph, runs tool calls (file edits, shell commands, web searches) for free, and manages context with smart compaction and verification. Features include a rich UI, slash commands, Git integration, sub‑agent delegation, extensive safety/permission checks, and reproducible benchmarks showing $0.027 per solved task at 23 s median time.

Crabwalk is a real‑time web UI that visualises OpenClaw AI agents’ activity across WhatsApp, Telegram, Discord and Slack. It streams events from the OpenClaw gateway via WebSocket, shows a live node‑graph of sessions, lets you inspect tool calls, and supports Docker, CLI, or source‑install deployment.

Agentica is an open‑source, Apache‑2.0‑licensed framework that lets you run multiple LLM‑powered agents locally. It provides a unified CLI, web UI, and desktop app that share the same history and configuration. Features include async‑first tool execution, multi‑model and multi‑modal support, persistent memory with context compression, a markdown‑based skill system that self‑evolves, safety guardrails, and built‑in collaboration (peer messaging, task/sub‑agent, and delegate processes). Python and TypeScript SDKs let you embed agents in code, and Docker/uv‑tool installers make setup easy.

A fast Python package for evaluating automatic speech recognition systems using metrics like Word Error Rate (WER) and Character Error Rate (CER).