* allow recognized callback to be called when LUIS response connectionMessage is empty
* fix response msg handler to account for bad intents
* Add test for bug
* Test stability fixes
* disable audio duration check due to service issus
* Increase translation timeout, format & add try / catch to synthesis test
* Change to a new version of Jest that supports retries. Modify tests to work with new version. Allow 3 retries.
* package-lock update
* Re-install with the same npm version as the ADO agents are using.
* Remove gulp-parent from forced resolution list
* npm audit auto fixes
* Increase error timeouts
* Increase timeouts
* Didn't copy the env change from the package.json to the jest config
* Add logging and increase token timeout.
Co-authored-by: Yulin Li <liyulin@pku.edu.cn>
* Initial implementation using logLevel alias for cleaner common imports
* Use LogLevel in tests
* lint bits
* Correct write file logic
* Add tests for Diagnostics
* Correct test name
* revert OCSP test changes
* Switch to use native promises inside the SDK, while keeping (most) external interfaces the same.
Visible Changes: (One breaking)
*PullAudioOutputStream.read() has a return type change from an internal Promise to a Native JavaScript Promise.
* ConversationTranslator.dispose()
* SpeakerAudioDestination.write()
* SpeakerAudioDestination.close()
* SpeakerAudioDestination.resume()
* BaseAudioPlayer.stop()
* SpeechSynthesizer.close()
* SpeechRecognizer.close()
* TranslationRecognizer.close()
* IntentRecognizer.close()
* Recognizer.close()
* DialogServiceConnector.connect()
* DialogServiceConnector.disconnect()
* Connection.openConnection()
* Connection.closeConnection()
Now all accept two callbacks as parameters (cb?: () => void, err?: (error: string) => void) to indicate status. These API's had never been fully synchronous and now status can be captured.
* Extend OCSP certificate verification to include a memory and disk based cache.
The disk cache will by default be located in TMPDIR/<username>/ but can be overridden.
When the cached OCSP response is within 1 day of expiring, or 1/2 way through it's vaility period (whichever is shorter) a background
task is queued to refresh it while the cached value is used.
* More fixes
* Fix error case
* Fix typos
* Remove throttling for the first 5 seconds of audio. After that throttle at 2x realtime.
* Update translation tests to reflect change in error message from service.
* Parallize opening of websocket connection with audio startup.
* Simplify recognize
* When an exception occurs, call the error callback and then dispose of the recognizer
* Adjust tests
* Push work
* Pull stream work
* On Chrome browsers, use Audio Worklet to capture audio
* Fix bugs merge introduced
* Add script URL code
* CR Feedback
* Disable empty test case
* Move intents to use phrase hints.
* Added classes & tests to base layer
* Initial add of PhraseListGrammar
* Add code comments and a couple of tests
* Improve comments
* Add support for Connection object.
Expose internalData object on base Recognizer class. This is used to allow the fromRecognizer* pattern on the connection object and other future objects that will be conjured from a recognizer.
Seperate the concept of a connection being established from one being configured with context information via the speech.config message.
Support multiple recognize* iterations being done in the same connection. Previously each recognizeOnce / startContinuousRecog call established a new connection.
Add & Update Tests.
* Add additional test
* Remove extra blank like
Glenn Harper
Ryan Hurey
Yulin Li
Microsoft Open Source