kstonekuan/tambourine-voice

What it solves

Tambourine provides a universal voice-to-text interface that allows users to dictate text into any application on their computer. It solves the limitations of native dictation tools by offering high personalization, the ability to remove filler words, and the flexibility to use various AI providers for transcription and formatting.

How it works

The system consists of a Tauri-based desktop application and a Python server. When a user triggers a hotkey, the app captures audio and streams it via WebRTC to the server. The server uses a Speech-to-Text (STT) provider to transcribe the audio and then passes that text through a Large Language Model (LLM) to clean up filler words, add punctuation, and apply custom formatting rules. The final polished text is then automatically typed back into the active application at the cursor position.

Who it’s for

It is designed for users who want to capture ideas faster than they can type, people who need a highly customizable voice interface for professional or technical workflows, and those who prefer open-source alternatives to proprietary dictation software.

Highlights

• Universal Compatibility: Works across emails, messages, documents, code editors, and terminals.
• Flexible AI Backend: Supports a wide range of STT (e.g., Deepgram, Groq, OpenAI) and LLM (e.g., Anthropic, Gemini, Cerebras) providers, including fully local options via Whisper and Ollama.
• Context-Aware Formatting: Automatically detects the focused application to tailor the output style (e.g., formal for emails, casual for messaging).
• Customizable Control: Users can define their own formatting prompts and maintain a personal dictionary for technical terms.
• Dual-Mode Recording: Offers both "hold-to-record" and "toggle" modes via global hotkeys.