New upgrade, new announcement, new feature: voice instruction recognition in solo sessions. Hurrah, this sounds really cool, and should indeed bring a lot of fun to some solo players... yet unfortunately a lot of frustration to others. Bear with me.
The feature relies on Sphinx thus requires a few extra dependencies to be enabled, but nothing breaks if they are not installed. See the updated README file for what to add to your system. All is free software and usable offline once installed of course. I hate to opt otherwise.
Once installed, ATC-pie should tell you "speech recognition modules found" at start-up, which confirms the feature will be available. If "not found", no point messing with the GUI options; fix your system/libraries first.
The fun part:
Once started with modules found, the general settings (now F11) solo tab offers new options including one for "voice instructions". Select it, use push-to-talk (Ctrl) in the game and speak without breaking, repeating, backtracking. Aircraft should comply if your instruction makes sense. A low beep (sound notif's on) means that the system did not recognise the instruction. Please read the quick reference and the updated wiki user guide.
The possibly frustrating facts:
- Installation for non-Linux users. For Linux, if nothing is funky about your computer or devices, everything you should need has been added to the README. Just go through the section and install the listed packets. For other platforms, I have no idea, and some testers have tried and failed but everything required is free and multi-platform software, so in theory everything is possible. Well-informed Windows/Mac users may know how best to transpose the dependencies and get them to work, and I am counting on them to write up a contribution to the README to help out their fellow platform users.
- Non-native American English accents. Strong accent variation is of course a potential problem. My (British) accent seems to get across quite fine, and several testers have managed to enjoy sessions while not being native, but their accents were mostly close enough. For the sake of inclusiveness, I have added an option to use a custom the acoustic model, so everybody can provide (train or adjust) one fitting their pronunciation. But, unless you have worked with speech recog before, you will not know what to do and may need some coffee before you try. No other solution, as far as I can tell.
I do hope you have fun with this. The feature to follow is of course pilot readback with speech synthesis.
[EDIT: missing word]