The chrome.experimental.speechInput module provides one-shot speech recognition to Chrome extensions. This module is still experimental. For information on how to use experimental APIs, see the chrome.experimental.* APIs page.

Manifest

You must declare the "experimental" permission in the extension manifest to use the speech input API. For example:

{
  "name": "My extension",
  ...
  "permissions": [
    "experimental"
  ],
  ...
}

How to start speech recognition

To start recognizing speech an extension must call the start() method. If provided, your callback will be called once recording has successfully started. In case of error chrome.extension.lastError will be set.

This API provides exclusive access to the default recording device to the first extension requesting it. Consequently, any calls to start() when the device is being used by another extension or web page will fail and set chrome.extension.lastError. The message requestDenied will be set if another extension in the same profile is making use of the API. Otherwise noRecordingDeviceFound, recordingDeviceInUse or unableToStart will be set depending on the situation.

To check whether recording is currently active, call the isRecording() method. Please note that it only checks for audio recording within Chrome.

How to get speech recognition results

Listen to the onResult event to receive speech recognition results.

var callback = function(result) { ... };

chrome.experimental.speechInput.onResult.addListener(callback);

The result object contains an array of SpeechInputResultHypothesis sorted by decreasing likelihood.

Recording automatically stops when results are received. It is safe to call start() again from the results callback.

To handle errors during speech recognition listen for the onError event.

var callback = function(error) { ... };

chrome.experimental.speechInput.onError.addListener(callback);

How to stop recording

To stop speech recognition call the stop() method. If provided, the callback function will be called once recording has successfully stopped. In case of error chrome.extension.lastError will be set.

Other features

• onSoundStart - Event generated when start of sound is detected (from previously being silent).
• onSoundEnd - Event generated when end of sound is detected (a continued period of silence).

Examples

The following example illustrates how to show a JavaScript alert with the most likely recognition result.

function checkStart() {
  if (chrome.extension.lastError) {
    alert("Couldn't start speech input: " + chrome.extension.lastError.message);
  }
}

function recognitionFailed(error) {
  alert("Speech input failed: " + error.code);
}

function recognitionSucceeded(result) {
  alert("Recognized '" + result.hypotheses[0].utterance + "' with confidence " + result.hypotheses[0].confidence);
}

chrome.experimental.speechInput.onError.addListener(recognitionFailed);
chrome.experimental.speechInput.onResult.addListener(recognitionSucceded);
chrome.experimental.speechInput.start({ "language": "en" }, checkStart);