Speech recognition library

readme.md

Library that we created: speech-to-text.min.js/speech-to-text.js. It's not open source. Its a wrapper that unify all APIs from different speech recognition services into one API.

How it works

0. If we gonna use azure/google we need to load some dependencies

<!-- For Azure -->
<script src="https://speech-poc.awsext.berlitz.net/lib/microsoft.cognitiveservices.speech.sdk.bundle.js"></script>

<!-- For Google -->
<script src="https://speech-poc.awsext.berlitz.net/lib/socket.js"></script>

1. Load our library and create a container:

<script src="https://speech-poc.awsext.berlitz.net/lib/speech-to-text.min.js"></script>

var speechToText = new SpeechToText.Container;

Container not functional yet, we need to add speech services that we gonna use (supported azure, google, browser for now). Wrappers for this services implemented by library.

2. Add speech service (one or all of them):

// Browser API service (chrome only):
speechToText.addService('browser', new SpeechToText.BrowserApiService({
    lang: 'en-US',
    continuous: true,
    interimResults: true
}));

// Azure service:
speechToText.addService('azure', new SpeechToText.AzureService({
    key: '45c5dfed064c4014860778115720b932', // or token: <token> (token need to be generated on backend service)
    region: 'westus',
    language: 'en-US'
}));

// Google service
speechToText.addService('google', new SpeechToText.GoogleService({
    socketUrl: 'https://speech-poc.awsext.berlitz.net:3556' // backend need to be implemented
}));

For example, if we use only azure, then only azure service need to be added.

3. Set service to use or services priority (services priority set as current service first available by order):

speechToText.setServicePriority(['azure', 'browser', 'google']);
//or
speechToText.setCurrentService('azure');

4. Now container is functional and can be used for speech recognition:

try {
    speechToText.init(); // check microphone, throw exception if none of services works
} catch (e) {
    console.log(e);
}

// speech recognition results callback
speechToText.onResults((text, isCorrect, isCorrectByAlternative, config, diffElement) => {
// do something with results
});

speechToText.onStopRecognition((config) => {
// callback that always fired, even if nothing is sayed to microphone
});

// Start recognition, this method can be attached to button onclick event
speechToText.startRecognition({answer: 'the man is holding a briefcase', alternatives: ['hello'], maxListenTime: 10});

Notes

1. Keys 'azure', 'google', 'browser' is custom string, so we can add multiple instances of one service with different configs, for example:

speechToText.addService('azure-1', new SpeechToText.AzureService({
    key: '45c5dfed064c4014860778115720b932',
    region: 'westus',
    language: 'en-US'
}));

speechToText.addService('azure-2', new SpeechToText.AzureService({
    key: 'some-other-key',
    region: 'westus',
    language: 'pt'
}));