Fixes

• callingresume on a pipeline without first callingpause no longer causes the pipeline to hang

This release includes minor fixes for the Azure speech recognizer.

Bug Fixes

• honor locale in azure speech recognizer
• add partial recognition listener for Azure
• avoid timeout for empty Azure transcripts

Features

• wakeword-only pipeline profile and empty ASR stage

Bug Fixes

• RefactoredgetOrDefault call in NLU module that caused crashes on devices running < API 24

Bug Fixes

• Reset wake word detector on pipeline deactivation to avoid spurious reactivation

Features

• New speech pipeline profile for using on-device wakeword detection and keyword recognition ASR simultaneously

Features

• Rasa NLU and dialogue policy
• Finalize prompts via Spokestack wrapper

Bug Fixes

• enforce ordering of TTS responses
• ensure pipeline is resumed when TTS is stopped withstopPlayback

Features

• The speech pipeline profile can now be set directly from theSpokestack builder object viawithPipelineProfile().
• Classes for a keyword model can be loaded from the model's metadata file instead of being specified manually (seekeyword-metadata-path).

Features

• added a new profile for using a keyword detector as ASR

Fixes

• a typo in the Spokestack ASR profile was preventing it from loading

This release updates a default configuration value of the keyword recognizer to match the current training regime for keyword models.