Version 3.8.1

Lots of changes since June! Summary below. Get all of these and more with a quick pip install --upgrade SpeechRecognition.

• Snowboy hotwords support for highly efficient, performant listening (thanks @beeedy!). This is implemented as the snowboy_configuration parameter of recognizer_instance.listen.
• Configurable Pocketsphinx models - you can now specify your own acoustic parameters, language model, and phoneme dictionary, using the language parameter of recognizer_instance.recognize_sphinx (thanks @frawau!).
• audio_data_instance.get_segment(start_ms=None, end_ms=None) is a new method that can be called on any AudioData instance to get a segment of the audio starting at start_ms and ending at end_ms. This is really useful when you want to get, say, only the first five seconds of some audio.
• The stopper function returned by listen_in_background now accepts one parameter, wait_for_stop (defaulting to True for backwards compatibility), which determines whether the function will wait for the background thread to fully shutdown before returning. One advantage is that if wait_for_stop is False, you can call the stopper function from any thread!
• New example, demonstrating how to simultaneously listen to and recognize speech with the threaded producer/consumer pattern: threaded_workers.py.
• Various improvements and bugfixes:
  • Python 3 style type annotations in library documentation.
  • recognize_google_cloud now uses the v1 rather than the beta API (thanks @oort7!).
  • recognize_google_cloud now returns timestamp info when the show_all parameter is True.
  • recognize_bing won't time out as often on credential requests, due to a longer default timeout.
  • recognize_google_cloud timeouts respect recognizer_instance.operation_timeout now (thanks @reefactor!).
  • Any recognizers using FLAC audio were broken inside Linux on Docker - this is now fixed (thanks @reefactor!).
  • Various documentation and lint fixes (thanks @josh-hernandez-exe!).
  • Lots of small build system improvements.