--- tags: - speech-to-text - automatic-speech-recognition - voice-input - asr - gguf - whisper - sensevoice - parakeet - qwen-asr language: - en - zh --- # Voconly **Voconly is a local-first AI voice input assistant for Windows.** Speak naturally and let AI turn your voice into text, then polish, translate, summarize, or structure it. Voconly combines **local speech recognition (ASR)** with optional **LLM processing**, making voice a more natural way to write, communicate, capture ideas, and interact with your computer. **Website:** https://www.voconly.com **GitHub:** https://github.com/xinkyle/Voconly --- ## What is Voconly? Voconly is an open-source **AI voice input and speech-to-text tool** designed for local, private, and flexible use. Instead of typing everything with a keyboard: **Press a hotkey → Speak → Transcribe → AI Process → Output** You can use Voconly anywhere you normally type — editors, browsers, email clients, chat applications, terminals, IDEs, and more. ### Key Features * **Local Speech-to-Text** — Run ASR models locally on your computer * **Real-time Transcription** — See your words as you speak * **AI Text Processing** — Polish, translate, rewrite, summarize, and structure * **Privacy First** — Process voice data locally without uploading recordings * **Offline Support** — Local ASR can work without an internet connection * **Global Hotkeys** — Trigger voice input from any application * **Flexible Models** — Choose different speech recognition and LLM models for different scenarios --- ## Open-Source Models This Hugging Face organization hosts **open-source models and resources used by Voconly**, with a focus on local speech recognition, voice AI, and efficient local inference. ### Speech Recognition / ASR * [Whisper Large V3 Turbo — GGUF](https://huggingface.co/voconly/whisper-large-v3-turbo-gguf) — Fast local speech-to-text * [SenseVoice Small](https://huggingface.co/voconly/sensevoice-small) — Multilingual speech recognition * [Qwen3-ASR 1.7B — GGUF](https://huggingface.co/voconly/Qwen3-ASR-1.7B-gguf) — Lightweight multilingual ASR * [Parakeet TDT 0.6B — GGUF](https://huggingface.co/voconly/parakeet-tdt-0.6b-v3-gguf) — Efficient streaming speech recognition * [Parakeet Unified EN 0.6B — GGUF](https://huggingface.co/voconly/parakeet-unified-en-0.6b-gguf) — English speech recognition * [Nemotron 3.5 ASR 0.6B — GGUF](https://huggingface.co/voconly/nemotron-3.5-asr-streaming-0.6b-gguf) — Streaming ASR * [Cohere Transcribe 2B — GGUF](https://huggingface.co/voconly/cohere-transcribe-03-2026-gguf) — Speech transcription ### Local LLMs * [Qwen3.5-9B Q4_K_M — GGUF](https://huggingface.co/voconly/Qwen3.5-9B-Q4_K_M-GGUF) — Efficient local LLM inference * [Qwen3-4B-Instruct Q4_K_M — GGUF](https://huggingface.co/voconly/Qwen3-4B-Instruct-2507-Q4_K_M-GGUF) — Lightweight instruction model **[Browse all Voconly models and resources](https://huggingface.co/voconly/models)** --- ## Why Local AI? Voconly is built around a simple idea: **Your voice should belong to you.** Voice input can contain private conversations, work information, ideas, and personal thoughts. That's why Voconly supports running speech recognition locally: * Your audio can stay on your computer * No cloud upload is required for local ASR * No dependency on online transcription quotas * Works offline when using local models * You choose which models and AI services to use --- ## How Voconly Uses These Models Voconly separates the voice input process into two stages: **Speech → ASR → Text → LLM → Final Output** ASR models handle speech recognition, while optional LLMs can transform the raw transcript into useful content. For example: > **Speak:** > “Help me organize this idea from earlier and get it ready to send to the client.“ > **ASR:** > Converts speech into text. ↓ > **LLM:** > Rewrites and structures the content. ↓ > **Output:** > A polished message ready to send. This architecture allows users to choose different models depending on their hardware, language, accuracy, speed, and privacy requirements. --- ## Use Cases Voconly can be used for: * **Content Creation** — Capture ideas, write articles, create scripts * **Work & Communication** — Emails, messages, reports, summaries * **Meetings** — Transcription and structured meeting notes * **Development** — Describe requirements, write Issues, explain technical ideas * **Learning** — Capture notes and learning insights * **Daily Journaling** — Record thoughts and ideas naturally --- ## Why Open Source? AI is changing who can create software. In the past, turning an idea into a product often required a team of designers, developers, and engineers. Today, one person can use AI to build things that previously required an entire team. Voconly is part of my exploration of this idea: > **What can one person + AI create?** I believe AI should not only make existing work faster. It should give more people the ability to **turn their ideas into real products**. Voconly is one small experiment in that direction. --- ## Learn More **Voconly Website:** https://www.voconly.com **GitHub Repository:** https://github.com/xinkyle/Voconly **Hugging Face Organization:** https://huggingface.co/voconly If Voconly is useful to you, consider giving the project a ⭐ **Star on GitHub**. --- ## About **幸勇（老幸.AI）** 15 years of entrepreneurship and long-term experience in AI, big data, and software products. Since 2024, I have been exploring AI coding and building software products with AI. Voconly is part of my ongoing exploration of **local AI, personal AI, and individual creativity in the AI era**. **One person. One idea. AI.**