TypeScript icon, indicating that this package has built-in type declarations

0.3.0 • Public • Published


package size vulnerabilities

A polyfill for the experimental browser Speech Recognition API which falls back to AWS Transcribe.


Note: this is not a polyfill for MediaDevices.getUserMedia() - check the support table in the link above.

Who is it for?

This Library is a good fit if you are already using AWS services (or you would just prefer to use AWS).

A polyfill also exists at: /antelow/speech-polyfill, which uses Azure Cognitive Services as a fallback. However, it seems to have gone stale with no updates for ~2 years.


  • An AWS account
  • A Cognito identity pool (unauthenticated or authenticated) with the TranscribeStreaming permission.

AWS Setup Guide

  1. In the AWS console, visit the Cognito section and click Manage Identity Pools.
  2. Click Create new identity pool and give it a name.
  3. To allow anyone who visits your app to use speech recognition (e.g. for public-facing web apps) check Enable access to unauthenticated identities
  4. If you want to configure authentication instead, do so now.
  5. Click Create Pool
  6. Choose or create a role for your users. If you are just using authenticated sessions, you are only interested in the second section. If you aren't sure what to do here, the default role is fine.
  7. Make sure your role has the TranscribeStreaming policy attached. To attach this to your role search for IAM -> Roles, find your role, click "Attach policies" and search for the TranscribeStreaming role.
  8. Go back to Cognito and find your identity pool. Click Edit identity pool in the top right and make a note of your Identity pool ID


Install with npm i --save speech-recognition-aws-polyfill

Import into your application:

import SpeechRecognitionPolyfill from 'speech-recognition-aws-polyfill'

Or use from the unpkg CDN:

<script src="https://unpkg.com/speech-recognition-aws-polyfill"></script>

Create a new instance of the polyfill:

const recognition = new SpeechRecognitionPolyfill({
  IdentityPoolId: 'eu-west-1:11111111-1111-1111-1111-1111111111', // your Identity Pool ID
  region: 'eu-west-1' // your AWS region

Alternatively, use the create method.

const SpeechRecognition = SpeechRecognititionPolyfill.create({
  IdentityPoolId: 'eu-west-1:11111111-1111-1111-1111-1111111111', // your Identity Pool ID
  region: "eu-west-1"

const recognition = new SpeechRecognition()

You can then interact with recognition the same as you would with an instance of window.SpeechRecognition

The recognizer will stop capturing if it doesn't detect speech for a period. You can also stop manually with the stop() method.

Support Table


Property Supported
lang Yes
grammars No
continuous Yes
interimResults No
maxAlternatives No
serviceURI No


Method Supported
abort Yes
start Yes
stop Yes


Events Supported
audiostart Yes
audioend Yes
start Yes
end Yes
error Yes
nomatch Yes
result Yes
soundstart Partial
soundend Partial
speechstart Partial
speechend Partial

Full Example

import SpeechRecognitionPolyfill from 'speech-recognition-aws-polyfill'

const recognition = new SpeechRecognitionPolyfill({
  IdentityPoolId: 'eu-west-1:11111111-1111-1111-1111-1111111111', // your Identity Pool ID
  region: 'eu-west-1' // your AWS region
recognition.lang = 'en-US'; // add this to the config above instead if you want

document.body.onclick = function() {

recognition.onresult = function(event) {
  const { transcript } = event.results[0][0]
  console.log('Heard: ', transcript)

recognition.onerror = console.error


  • Further increase parity between the two implementations by better supporting additional options and events.
  • Build a companion polyfill for speech synthesis (TTS) using AWS Polly
  • Provide a way to output the transcription as an RxJS observable

Contributing and Bugs

Questions, comments and contributions are very welcome. Just raise an Issue/PR (or, check out the fancy new Github Discussions feature)



Package Sidebar


npm i speech-recognition-aws-polyfill-react

Weekly Downloads






Unpacked Size

9.3 MB

Total Files


Last publish


  • ferres