Support overriding recognition mode for continuous speech recognition about cognitive-services-speech-sdk-js

[rhurey]

The data type for the endpoint isn't of the expected type.

fromEndPoint is defined as:
public static fromEndpoint(endpoint: URL, subscriptionKey: string): SpeechConfig

So in this case, you'd want something like: (node)
var URL = require('url').URL;
let speechConfig = sdk.SpeechConfig.fromEndpoint(new URL (wss://${SERVICE_REGION}.stt.speech.microsoft.com/speech/recognition/dictation/cognitiveservices/v1), SUBSCRIPTION_KEY);

The SDK's parameter check at that point was only for null on the passed in parameter, it didn't type check the parameter or the .href property it used. I'll put a bug in our backlog to tighten the parameter check and get a better error message out.

Hope that helps.

[orgads]

ping?

[wolfma61]

currently not possible by setting a property. We are going to try to address this in an upcoming release.

as a work-around you can create the recognizer fromEndpoint
https://docs.microsoft.com/en-us/javascript/api/microsoft-cognitiveservices-speech-sdk/speechconfig?view=azure-node-latest#fromendpoint-url--string-

the URL should look similar to this:

wss://.stt.speech.microsoft.com/speech/recognition/dictation/cognitiveservices/v1

[orgads]

Thank you. I'll try it.

[geooot]

Hello!
I am trying to use dictation mode as well and this is how I setup the speechConfig using the workaround. It worked using fromSubscription, but doesn't work using fromEndpoint. I am kind of guessing on how the url should be formatted as I have not found a reference to a websocket url in the azure portal.
If you could see what I am doing wrong, help would be much appreciated

// now create the audio-config pointing to our stream and
// the speech config specifying the language.
let audioConfig = sdk.AudioConfig.fromStreamInput(pushStream);
// let speechConfig = sdk.SpeechConfig.fromSubscription(SUBSCRIPTION_KEY, SERVICE_REGION);
let speechConfig = sdk.SpeechConfig.fromEndpoint(`wss://${SERVICE_REGION}.stt.speech.microsoft.com/speech/recognition/dictation/cognitiveservices/v1`, SUBSCRIPTION_KEY);

// setting the recognition language to English.
speechConfig.speechRecognitionLanguage = LANG;

// create the speech recognizer.
let recognizer = new sdk.SpeechRecognizer(speechConfig, audioConfig);

[orgads]

I've noticed that fromEndpoint uses subscription key. Is there an alternative that accepts token (and still enabled dictation mode)? I want to avoid issueToken for each client.

[rhurey]

You should be able to pass in any value you want for the subscription key passed to fromEnedPoint and then call recognizer.authorizationToken = and the token will be used instead of the subscription key.

[rhurey]

enableDictation was added to the SpeechConfig object in the 1.7.0 release.