I'm using Azure SpeechSDK services for speech-to-text transcription using recognizeOnceAsync
. The current code resembles:
var SpeechSDK, recognizer, synthesizer;
var speechConfig = SpeechSDK.SpeechConfig.fromSubscription('SUB_KEY', 'SUB_REGION');
var audioConfig = SpeechSDK.AudioConfig.fromDefaultMicrophoneInput();
recognizer = new SpeechSDK.SpeechRecognizer(speechConfig, audioConfig);
new Promise(function(resolve) {
recognizer.onend = resolve;
recognizer.recognizeOnceAsync(
function (result) {
recognizer.close();
recognizer = undefined;
resolve(result.text);
},
function (err) {
alert(err);
recognizer.close();
recognizer = undefined;
}
);
}).then(r => {
console.log(`Azure STT enterpreted: ${r}`);
});
In the HTML file I import the Azure package like so:
<script src="https://aka.ms/csspeech/jsbrowserpackageraw"></script>
The issue is that I would like to increase the amount of "Silence time" which is allowed before the recognizeOnceAsync
method returns the result. (I.e. you should be able to stop and take a breath without the method assuming you're done talking). Is there any way to do this with fromDefaultMicrophoneInput
? I've tried various things like:
const SILENCE_UNTIL_TIMEOUT_MS = 5000;
speechConfig.SpeechServiceConnection_EndSilenceTimeoutMs = SILENCE_UNTIL_TIMEOUT_MS;
audioConfig.setProperty("Speech_SegmentationSilenceTimeoutMs", SILENCE_UNTIL_TIMEOUT_MS);
but none seem to extend the "silence time allowance" correctly.
This is the resource which I have been looking at: https://docs.microsoft.com/en-us/javascript/api/microsoft-cognitiveservices-speech-sdk/propertyid?view=azure-node-latest
CodePudding user response:
Based on what you're describing, you'd need to set the segmentation silence timeout. Unfortunately, there is a bug in the JS SDK at the moment and the PropertyId.Speech_SegmentationSilenceTimeoutMs
is not being set correctly.
As a workaround, you can instead set the segmentation timeout as follows:
const speechConfig = SpeechConfig.fromSubscription(subscriptionKey, subscriptionRegion);
speechConfig.speechRecognitionLanguage = "en-US";
const reco = new SpeechRecognizer(speechConfig);
const conn = Connection.fromRecognizer(reco);
conn.setMessageProperty("speech.context", "phraseDetection", {
"INTERACTIVE": {
"segmentation": {
"mode": "custom",
"segmentationSilenceTimeoutMs": 5000
}
},
mode: "Interactive"
});
reco.recognizeOnceAsync(
(result) =>
{
console.log("Recognition done!!!");
// do something with the recognition
},
(error) =>
{
console.log("Recognition failed. Error:" error);
});
Please note that the allowed range for the segmentation timeout is 100-5000 ms (inclusive)