Overview

An extension can register itself as a speech engine. By doing so, it can intercept some or all calls to functions such as $ref:tts.speak and $ref:tts.stop and provide an alternate implementation. Extensions are free to use any available web technology to provide speech, including streaming audio from a server, HTML5 audio, Native Client, or Flash. An extension could even do something different with the utterances, like display closed captions in a pop-up window or send them as log messages to a remote server.

Manifest

To implement a TTS engine, an extension must declare the "ttsEngine" permission and then declare all voices it provides in the extension manifest, like this:

{
  "name": "My TTS Engine",
  "version": "1.0",
  "permissions": ["ttsEngine"],
  "tts_engine": {
    "voices": [
      {
        "voice_name": "Alice",
        "lang": "en-US",
        "gender": "female",
        "event_types": ["start", "marker", "end"]
      },
      {
        "voice_name": "Pat",
        "lang": "en-US",
        "event_types": ["end"]
      }
    ]
  },
  "background": {
    "page": "background.html",
    "persistent": false
  }
}

An extension can specify any number of voices.

The voice_name parameter is required. The name should be descriptive enough that it identifies the name of the voice and the engine used. In the unlikely event that two extensions register voices with the same name, a client can specify the ID of the extension that should do the synthesis.

The gender parameter is optional. If your voice corresponds to a male or female voice, you can use this parameter to help clients choose the most appropriate voice for their application.

The lang parameter is optional, but highly recommended. Almost always, a voice can synthesize speech in just a single language. When an engine supports more than one language, it can easily register a separate voice for each language. Under rare circumstances where a single voice can handle more than one language, it's easiest to just list two separate voices and handle them using the same logic internally. However, if you want to create a voice that will handle utterances in any language, leave out the lang parameter from your extension's manifest.

Finally, the event_types parameter is required if the engine can send events to update the client on the progress of speech synthesis. At a minimum, supporting the 'end' event type to indicate when speech is finished is highly recommended, otherwise Chrome cannot schedule queued utterances.

Note: If your TTS engine does not support the 'end' event type, Chrome cannot queue utterances because it has no way of knowing when your utterance has finished. To help mitigate this, Chrome passes an additional boolean enqueue option to your engine's onSpeak handler, giving you the option of implementing your own queueing. This is discouraged because then clients are unable to queue utterances that should get spoken by different speech engines.

The possible event types that you can send correspond to the event types that the speak() method receives:

• 'start': The engine has started speaking the utterance.
• 'word': A word boundary was reached. Use event.charIndex to determine the current speech position.
• 'sentence': A sentence boundary was reached. Use event.charIndex to determine the current speech position.
• 'marker': An SSML marker was reached. Use event.charIndex to determine the current speech position.
• 'end': The engine has finished speaking the utterance.
• 'error': An engine-specific error occurred and this utterance cannot be spoken. Pass more information in event.errorMessage.

The 'interrupted' and 'cancelled' events are not sent by the speech engine; they are generated automatically by Chrome.

Text-to-speech clients can get the voice information from your extension's manifest by calling $ref:tts.getVoices, assuming you've registered speech event listeners as described below.

Handling speech events

To generate speech at the request of clients, your extension must register listeners for both onSpeak and onStop, like this:

var speakListener = function(utterance, options, sendTtsEvent) {
  sendTtsEvent({'event_type': 'start', 'charIndex': 0})

  // (start speaking)

  sendTtsEvent({'event_type': 'end', 'charIndex': utterance.length})
};

var stopListener = function() {
  // (stop all speech)
};

chrome.ttsEngine.onSpeak.addListener(speakListener);
chrome.ttsEngine.onStop.addListener(stopListener);

Important: If your extension does not register listeners for both onSpeak and onStop, it will not intercept any speech calls, regardless of what is in the manifest.

The decision of whether or not to send a given speech request to an extension is based solely on whether the extension supports the given voice parameters in its manifest and has registered listeners for onSpeak and onStop. In other words, there's no way for an extension to receive a speech request and dynamically decide whether to handle it.