AudioWizard
Listen instead of read. Neural text-to-speech for Markdown docs, long specs and AI agent replies, right inside VS Code, Cursor and Antigravity.
The Problem
When working with AI coding agents (Claude Code, Gemini, Copilot) or reviewing long architectural specs, planning docs, and multi-thousand-line context files, developers suffer from severe visual fatigue. Most agents do not speak their long answers, and screen readers fail by reading raw code syntax, semicolons, and curly braces verbatim.
AudioWizard gives you a speaker button for all of it: free, with natural neural voices and no API key.
How to use it
| Do this |
Press |
| Read the current file (from the cursor, or from the top) |
Ctrl+Alt+R · ⌘⌥R |
| Read your selection, or an agent reply you selected or copied |
Ctrl+Alt+C · ⌘⌥C |
| Read the latest Claude Code reply, without copying anything |
Ctrl+Alt+A · ⌘⌥A |
| Play / pause |
Ctrl+Alt+Space · ⌘⌥P |
| Stop |
Ctrl+Alt+X · ⌘⌥X (or Esc in the player) |
You can also click the speaker icon in the title bar of any Markdown or text file.
Agent replies:
Claude Code: press Ctrl+Alt+A (or the 💬 button in the player) to hear its latest finished reply for the project you have open. If Claude Code is still working, you hear the previous answer.
Any other agent: select text in its chat panel and press Ctrl+Alt+C. If a panel doesn't let AudioWizard copy the selection, click the reply's Copy button first, then press Ctrl+Alt+C.
Antigravity's Agent panel: turn on Read copies in the player (or run AudioWizard: Toggle Read What I Copy). Then select text in the Agent panel and press Ctrl+C, or click a reply's Copy button, and it's read aloud. Antigravity's panel doesn't reliably pass Ctrl+Alt+… shortcuts on to extensions and keeps its conversations encrypted, so Ctrl+Alt+C and Ctrl+Alt+A don't work there. Use the player's buttons to pause or stop.
Read copies only reads text copied while the editor window is focused, skips copies from your own text editor, and skips single words or tokens, so passwords copied in other apps are never read.
The player
AudioWizard opens a compact player in the bottom panel, next to the Terminal. You can drag it to a sidebar if you prefer.
- Play / pause, previous / next sentence, and a progress bar you can click to jump.
- The text, with the current sentence highlighted. Click any sentence to start reading from there.
- Speed from 0.75× to 2.5×, and 13 neural voices (US, UK, India, Australia).
- Full or Skim. Skim reads the headings and the first sentence of every paragraph and list item, so you hear every point of a long doc without the detail.
Long files start speaking in a second or two: AudioWizard reads sentence by sentence and prepares the next ones while you listen.
Reads Markdown like a person would
- Code blocks are announced, not read: "Code block, 12 lines, skipped."
- Tables become spoken lists, links read their text, and URLs read as "link to github.com".
- Headings, bullets, bold and checkboxes are read as plain text, and
snake_case names stay intact.
Settings
| Setting |
Default |
What it does |
audiowizard.engine |
neural |
neural: Microsoft Edge neural voices. system: your operating system's voices, fully offline. |
audiowizard.voice |
en-US-JennyNeural |
Neural voice. Also changeable in the player. |
audiowizard.speed |
1 |
0.5× to 2.5×. Also changeable in the player. |
audiowizard.skipCodeBlocks |
true |
Announce code blocks instead of reading code aloud. |
audiowizard.autoRead |
false |
Pro. Read each Claude Code reply aloud as soon as it finishes. |
audiowizard.ai.provider |
auto |
Pro. Which AI writes Overviews. auto uses Claude Code when it's installed, otherwise Gemini. Also claude-code, gemini, anthropic, openai, ollama. |
audiowizard.ai.model |
(empty) |
Pro. Leave empty for the default (haiku for Claude Code, gemini-3.8-flash, claude-opus-5-5, gpt-6-luna, llama3.2). |
audiowizard.ai.claudePath |
(empty) |
Pro. The claude program for Overview. Empty: the one that comes with the Claude Code extension, or one on your PATH. |
audiowizard.ai.ollamaUrl |
http://localhost:11434 |
Pro. Your Ollama server. |
Privacy
With neural voices, the text being read is sent, sentence by sentence, to Microsoft's Edge speech service to generate the audio. Nothing else from your project is sent. Switch audiowizard.engine to system and nothing leaves your machine. No accounts, no analytics.
Claude Code replies are read from the conversation files Claude Code keeps on your computer; AudioWizard never uploads them. Overview (Pro) sends the text you asked for to the AI you chose: your own Claude Code (under your Claude sign-in), a provider you gave a key for, or Ollama on your own machine. Details: PRIVACY.md.
The neural voices come from Microsoft's Edge read-aloud service, which isn't an official API and can change. If it can't be reached, AudioWizard switches to your system voice automatically.
AudioWizard Pro: $4.99 once
No subscription. Everything above stays free.
- Auto-read. Turn on the switch in the player and every Claude Code reply is read aloud the moment it finishes. Ask, lean back, listen.
- Overview. A one-minute spoken summary of a long doc or agent reply. Using Claude Code? There's nothing to set up: AudioWizard asks your own Claude Code to write it, on your Claude plan, with no API key. Not on Claude Code? Bring a key for Google Gemini (free keys available), Anthropic or OpenAI, or use a local Ollama model with no key and nothing leaving your machine. Pick Overview in the player, or run AudioWizard: Overview of File. Overview and Auto-read work together: hear a summary of each reply.
Get Pro with AudioWizard: Get AudioWizard Pro, then AudioWizard: Activate Pro License with the key from your receipt. To pick a different AI or set a key, run AudioWizard: Choose AI for Overview; keys are kept in your editor's secret storage. Pro license terms.
Known limitations
- On keyboard layouts that use AltGr (Polish, German and others),
Ctrl+Alt shortcuts can clash with typing special characters. Rebind them in Keyboard Shortcuts (search "AudioWizard").
- Reading highlighted chat text depends on the chat panel supporting the editor's Copy command. The Copy-button route always works.
Development
npm install
npm test # parser tests
npm run build # bundle to dist/
npm run package # typecheck, test, build and create the .vsix
License
MIT © Dr. OpenSource
| |
