annyang

More forgiving voice commands

Speech recognition sometimes adds punctuation to what you say. annyang now handles common sentence punctuation automatically, with no new options to configure.

annyang.addCommands({
  'hello there': () => { console.log('Hello!'); },
});

// All three match the same command:
annyang.trigger('hello there');
annyang.trigger('Hello there!');
annyang.trigger('Hello, there.');

Commands written with punctuation, like 'Hello, there!', work too.

New Features

• More forgiving punctuation matching — String commands tolerate sentence-ending periods, exclamation marks, and question marks after literal text, and commas between literal words. Captured values and custom regular expressions are unchanged.

Bug Fixes

• Unicode case-insensitive matching — String commands now correctly match uppercase and lowercase letters outside the basic multilingual plane, including Deseret and Osage.

Internal

• Updated development dependencies to address security advisories.
• Kept the patched humanfs dependency compatible with Node.js versions below 24.

Rewritten in TypeScript, modern modules, and more!

annyang v3 is a ground-up rewrite. The library is now written in TypeScript, ships its own type definitions, and supports every module format — ESM, CommonJS, and IIFE script tags. It's still tiny (2 KB), still has zero dependencies, and now works safely in browsers that don't support speech recognition.

import annyang from 'annyang';

annyang.addCommands({
  'search for *term': (term) => { console.log(`Searching for ${term}`); },
});

if (annyang.isSpeechRecognitionSupported()) {
  annyang.start();
} else {
  // trigger() works even without speech recognition —
  // great for testing, automation, or fallback UIs in non-supporting browsers
  annyang.trigger('search for cat videos');
}

Named imports work too:

import { addCommands, start, trigger, isSpeechRecognitionSupported } from 'annyang';

New Features / Breaking Changes

• TypeScript types included — Full type definitions ship with the package.
• addCallback enforces correct callback signatures per event type.
• ESM/CJS/IIFE module support — Works with import, require(), and <script> tags.
• getState() method — Returns 'idle', 'listening', or 'paused'.
• state property (on default export) — Getter that returns the current state.
• addCallback returns an unsubscribe function:

  const unsub = annyang.addCallback('start', myFunc);
  unsub(); // removes myFunc from 'start' callbacks
  

• trigger() works independently of speech recognition — trigger() can now be used regardless of whether annyang is listening, paused, aborted, or even in browsers that don't support speech recognition.
• Safe to use in unsupported browsers — Methods like start(), addCommands(), and setLanguage() no longer throw when SpeechRecognition is unavailable.
• if (annyang) no longer detects browser support — The annyang object is always defined in v3. Use isSpeechRecognitionSupported() instead:

  // v2
  if (annyang) {
    annyang.start();
  }
  
  // v3
  if (annyang.isSpeechRecognitionSupported()) {
    annyang.start();
  }
  

• init() deprecated — annyang has been initializing automatically since v1.1.0. Calling init() now logs a deprecation warning. Remove any calls to init().
• String-based command callbacks removed — Passing function names as strings (e.g., {'hello': 'sayHello'}) is no longer supported. Pass functions directly: {'hello': sayHello}.
• Duplicate command phrases now overwrite — In v2, adding a command with the same phrase would register both. In v3, the new callback replaces the old one.

Internal

• Source rewritten in TypeScript with strict mode
• Bundler switched from Rollup to tsup
• Tests migrated to Vitest

👍 This release is 100% backwards compatible with all v2 releases.

Ensured support up to Node.js 10. Updated dependencies and rebuilt for a more efficient distributable.

Awesome people :star:

Special thanks to:

• @DanielRuf
• @Gebrukal
• @HannesOberreiter

👍 This release is 100% backwards compatible with all v2 releases.

[Added] Allow listening to the soundstart event

You can now add a callback event which will run when the speech recognition's soundstart event fires.

annyang.addCallback('soundstart', fn);

This event will fire once per speech recognition session (due to a bug in Chrome). If you would like to get it again the next time sound is heard, you can abort and restart speech recognition… which is the default behavior in HTTPS anyway. In other words, in HTTPS it works great.

See a more complete example for detecting when the user starts and stops speaking in the FAQ.

Changed file structure and development stack

The main annyang files which you will include in your project are now in the /dist directory:

• /dist/annyang.js
• /dist/annyang.min.js

If you would like to work on annyang's code itself, it is now in the /src directory:

• /src/annyang.js

annyang's source now passes through babel with the es2015 preset, so you can use ES2015 language features in it.

Awesome people :star:

Special thanks to:

• @remybach - For the original soundstart event code, and helping determine the final functionality.
• @peterdillon - Helping create a use case to test the new soundstart event.