中文版 | English

Free, open-source, system-wide voice-to-text for Ubuntu. Press PgDn to record, press again to transcribe, and VibeMic pastes the text into your current app. Works in any X11 application.

git clone https://github.com/ithiria894/vibemic-native-ubuntu.git
cd vibemic-native-ubuntu
chmod +x setup.sh
./setup.sh
python3 vibemic.py

On first launch, VibeMic detects your hardware and walks you through setup. No manual configuration needed.

VibeMic works with Groq, which offers a free Whisper API with generous limits:

• ~8 hours of transcription per day, no credit card required
• Runs whisper-large-v3-turbo on Groq's LPU hardware, so results come back in under a second
• Sign up at console.groq.com and create an API key at console.groq.com/keys

If you don't have a GPU, Groq is the recommended backend. VibeMic's first-launch wizard will guide you through it.

• System-wide -- works in any X11 app, not just editors
• One-key toggle -- PgDn starts and stops recording
• Instant paste -- clipboard + Ctrl+V, no per-character typing delay
• First-launch wizard -- auto-detects GPU and recommends the right backend
• Multiple providers -- OpenAI, Groq (free), LiteLLM, Custom OpenAI-compatible, Local whisper.cpp
• Per-provider settings -- each provider remembers its own API key and model
• Transcript history -- browse, copy, and delete past transcripts from the tray menu
• Native settings -- clean card-based UI, configure everything from the tray icon
• Local model library -- download Whisper models directly from Settings
• Paraphrase mode -- optional AI rewrite after transcription
• Error logging -- all errors logged to ~/.local/share/vibemic/vibemic.log
• Desktop notifications -- native Ubuntu notifications for recording state

If you have a GPU (NVIDIA CUDA, AMD Vulkan, or Intel Vulkan), local transcription is fast and private:

Without a GPU, local transcription on CPU is slow (~50 seconds per short clip). Use Groq instead.

• Ubuntu 20.04+ (or any Linux with X11)
• Python 3.8+
• sox for audio recording
• xdotool + xclip for clipboard paste
• python3-xlib for X11-level hotkey grab (optional, falls back to pynput)
• For Groq/OpenAI: an API key (Groq is free)
• For local: whisper-cli from whisper.cpp

sudo apt install sox libsox-fmt-all xdotool xclip libnotify-bin python3-tk python3-xlib
pip3 install --user openai pystray pynput Pillow
python3 vibemic.py

Right-click the tray icon and choose Settings. The UI is grouped into cards:

Transcription -- pick your provider, enter API key, select model. Each provider remembers its own key and model separately.

Recognition -- language hint, vocabulary hint for domain-specific words, response format (OpenAI/Custom only).

Paraphrase -- optional AI rewrite pass after transcription. Uses the same API key.

Preferences -- record hotkey (default: PgDn).

Settings are saved to config.json. Errors are logged to ~/.local/share/vibemic/vibemic.log.

Text is copied to the clipboard with xclip and pasted using xdotool key ctrl+v. This is fast even for long text and works well with CJK characters.

• VibeMic Native macOS

MIT

免費開源嘅 Ubuntu 全系統語音轉文字工具。按 PgDn 開始錄音，再按一次就轉錄並貼上到任何應用程式。

git clone https://github.com/ithiria894/vibemic-native-ubuntu.git
cd vibemic-native-ubuntu
chmod +x setup.sh
./setup.sh
python3 vibemic.py

第一次啟動會自動偵測你嘅硬件，引導你完成設定。

Groq 提供免費嘅 Whisper API：

• 每日約 8 小時免費轉錄，唔使信用卡
• 用 whisper-large-v3-turbo，結果一秒內返回
• 去 console.groq.com/keys 申請免費 API key

冇 GPU 嘅話，Groq 係最推薦嘅選擇。

• 全系統使用，任何 X11 app 都得
• 一鍵錄音 / 停止（PgDn）
• 剪貼簿即時貼上
• 首次啟動 wizard（自動偵測 GPU，推薦最適合嘅 backend）
• 多個 provider：OpenAI、Groq（免費）、LiteLLM、Custom、Local whisper.cpp
• 每個 provider 獨立記住 API key 同 model
• Tray menu 內建 history
• 原生 card-based 設定 UI
• 本機 model library 直接下載
• AI 改寫模式（Paraphrase）
• 錯誤自動記錄到 ~/.local/share/vibemic/vibemic.log

如果冇 GPU，用 Groq（免費、快）。

如果有 GPU，用 Local whisper.cpp + Fast - Large v3 Turbo Q5（離線、私密）。

如果主要講廣東話 + 英文夾講，model 揀 Recommended - Large v3 Q5。