drmfinlay / sphinxwrapper Goto Github PK

The upstream project (Pocket Sphinx) was updated recently, so I'd like to update this project in turn. The notable changes to be made are as follows:

• Remove the old extension code altogether, since it is now quite out-of-date.
  It would be nice to make use of it, but I don't see much point.
• Release a new version of this package that requires a version of the pocketsphinx package with binaries for Python 2.7.
  This is to finalise the sphinxwrapper package and Python 2.7 support.
• Change the project name from "sphinxwrapper" to "simplepocketsphinx".
  This name is more descriptive; the callback approach used in this project is simpler if one needs to do processing when speech begins and when it ends.
• Release a new version of this package under the simplepocketsphinx name.

Since there have been many breaking changes to Pocket Sphinx, I will not be updating the project to use version 5. I may do this at some point in the future.

I've started a new project to allow the standard US English CMU Pocket Sphinx models and dictionaries to be used with non-US variants of English by using GNU Aspell to localise words. The idea is to be able to write grammars and have key phrase searches in say, Australian English, and receive output words in Australian English without having to make your own model(s).

This functionality will be optional through use of the setuptools extras_require parameter in order to keep sphinxwrapper portable. So if the aspell package isn't available, everything will still work.

If you want to use sphinxwrapper with other languages, this could provide spell checking support for grammar / key phrase searches if the language is supported by GNU aspell and you have the dictionary installed.

I've come across this rather unhelpful error that gets raised sometimes:

Traceback (most recent call last):
  File "examples/recognise_forever.py", line 28, in <module>
    main()
  File "examples/recognise_forever.py", line 25, in main
    time.sleep(0.1)
TypeError: expected string or Unicode object, NoneType found

I've commented out time.sleep(0.1) in the examples for now which stops this from happening, but it isn't ideal because we can't stop hogging the CPU from Python with this bug.

I believe it has something to do with the multi-threaded audio implementations in sphinxbase. I've tried manipulating Python's Global Interpreter Lock following the Python 2.7 docs on this, but either I'm not doing it correctly or the issue is elsewhere. I know that this didn't happen before commit c344210.

There are docstrings for everything available through the Python help function using help(sphinxwrapper) or similar in the interpreter. I'm not quite sure how to get a documentation engine like Sphinx (lol) to pull the Python docstrings in the module into an easy to access format like HTML or markdown.