TIA Voice

TIA Voice is an open source, context-aware voice assistant for desktop. Hold a global hotkey, speak, and your words are transcribed, intelligently refined, and pasted directly into the app you are using. It also provides meeting capture, live captions, text-to-speech playback, and a smart selection toolbar — making it more than just a dictation tool.

Screenshots

Dashboard	Meeting capture

Meeting detail	Live Caption

Comparison with Similar Tools

TIA Voice is inspired by desktop voice dictation tools like MacWhisper, Wispr Flow, and Superwhisper, but goes beyond pure dictation with several unique capabilities:

Feature	TIA Voice	MacWhisper	Wispr Flow	Superwhisper
Push-to-talk global dictation	✅	✅	✅	✅
LLM-powered cleanup & rewrite	✅	❌	✅	✅
Custom post-process presets	✅	❌	❌	❌
Dictionary normalization	✅	❌	❌	❌
Intent routing (dictate vs. edit vs. Q&A)	✅	❌	❌	❌
Meeting capture with summaries	✅	❌	❌	❌
Live system-audio captions & translation	✅	❌	❌	❌
Selection toolbar with TTS	✅	❌	❌	❌
Text-to-speech with word highlighting	✅	❌	❌	❌
Multi-provider (DashScope / OpenAI)	✅	✅ (local)	✅	✅
BYO API key, data stays local	✅	N/A	✅	✅
Open source	✅	❌	❌	❌

What It Does

Voice Dictation & Smart Paste

Hold the push-to-talk key (Right Command on macOS / Right Alt on Windows), speak naturally, and release. TIA Voice transcribes your speech, cleans it up with an LLM, and pastes the result wherever your cursor is.

Intelligent Intent Routing

TIA Voice understands context and adapts its behavior automatically:

• Dictation mode — No text selected, cursor in a text field: transcribe and paste your spoken words.
• Edit mode — Text is selected in a text field: your voice command rewrites the selected text (e.g., select a sentence and say "make this more formal").
• Q&A mode — Text is selected outside a text field (e.g., in a browser): your voice question is answered based on the selected text.

Meeting Capture

Press Control+R to capture a meeting. TIA Voice records microphone audio and system audio, streams both through DashScope Gummy realtime transcription, and keeps speakers separated as You and Others.

• Saves a mixed local meeting audio file and raw transcript.
• Automatically generates a title, polished transcript, and meeting summary after capture ends.
• Stores meeting history locally with playback, summary, polished transcript, and raw transcript views.

Live Caption

Press Control+L to open Live Caption for system audio. TIA Voice shows a compact always-on-top caption overlay for calls, videos, or any audio playing on the machine.

• Supports auto-detected source captions plus optional translation.
• Can show the original text below translated captions.
• Shares the same system-audio transcription path used by meeting capture, and can also mirror the Others stream during a meeting.

Selection Toolbar

When you select text in your browser, a floating toolbar appears with instant actions:

• Read Out Loud — Converts selected text to natural-sounding speech via CosyVoice TTS, with a playback window that highlights each word in sync with the audio.

Text-to-Speech Player

Beyond the selection toolbar, TIA Voice includes a full TTS player:

• Powered by Alibaba DashScope CosyVoice v3 for high-quality, natural speech.
• Word-level timestamp synchronization — the transcription highlights word-by-word as audio plays.
• Play/pause, seek, and progress controls in a compact floating window.

LLM Post-Processing & Presets

Your spoken words pass through an LLM for intelligent cleanup:

• Fixes punctuation, grammar, and natural phrasing without altering meaning.
• Built-in presets: Formal (professional tone) and Casual (conversational tone).
• Custom presets: Define your own system prompts for specific writing styles.
• Toggle post-processing on or off per preset.

Dictionary Normalization

Define phrase mappings (e.g., "Buildmind" → "BuildMind") to automatically normalize commonly mis-transcribed terms. Dictionary entries are injected as high-priority rules into the LLM prompt.

Usage Statistics

Track your voice usage from the home dashboard: total words spoken, average words per minute, and transcription count with a scrollable history.

Multi-Provider Support

Choose your AI backend:

• ASR (Speech-to-Text): DashScope Qwen ASR Flash / OpenAI Whisper
• Realtime meeting/caption transcription: DashScope Gummy
• LLM (Cleanup & Intent): DashScope Qwen3.5 Flash / OpenAI GPT
• TTS (Text-to-Speech): DashScope CosyVoice v3

Bring your own API key — your data is processed directly through your provider and never touches a third-party server.

Setup

pnpm install
pnpm dev

On first launch:

1. Enter your DashScope (or OpenAI) API key in the onboarding dialog.
2. Grant macOS Accessibility permission when prompted (required for global hotkey and paste).
3. Start dictating with the default push-to-talk shortcut.

Development

pnpm dev          # Start in development mode
pnpm test:run     # Run tests
pnpm typecheck    # Type-check the project
pnpm lint         # Lint the project

Build

pnpm build        # Build for current platform
pnpm build:mac    # Build macOS distributable
pnpm build:win    # Build Windows distributable
pnpm build:linux  # Build Linux distributable

Tech Stack

• Runtime: Electron + React + TypeScript
• Styling: Tailwind CSS + shadcn/ui + Radix UI
• AI SDK: Vercel AI SDK (ai package)
• Global Hotkeys: uiohook-napi (native)
• Clipboard & Paste: @nut-tree-fork/nut-js
• Text Selection Hook: selection-hook (native, Chrome-based browsers)
• Realtime Meeting/Captions: DashScope Gummy with Electron system-audio capture
• TTS: DashScope CosyVoice API with word-level timestamps

Notes

• The default push-to-talk key is Right Command on macOS and Right Alt on Windows. You can change it to Right Option or Right Control in Settings.
• Meeting capture uses Control+R; Live Caption uses Control+L.
• System-audio capture on macOS works best on macOS 14.2+. Older macOS versions may need a virtual audio device for system audio.
• DashScope requests use https://dashscope.aliyuncs.com/compatible-mode/v1 by default. Override with DASHSCOPE_BASE_URL if you need a proxy.
• Your API key is stored locally in the app settings on the current machine and never leaves your device except for direct API calls.
• uiohook-napi is a native dependency. If the global hotkey fails to initialize after install, run pnpm rebuild uiohook-napi and restart the app.