I am extraordinarily happy to announce that OpenEars 1.0 is out now. This version has so many fixes and new features:
- Ultra-fast, ultra-easy install
- All-C language modelling and 3x faster phonetic lookup fallback method for vastly faster dynamic language model generation
- Language model generation is now tuned to creating contextual models from long sentences with improved recognition accuracy
- Only runs with ios 4 and 5 since the only devices that need 3 can’t use OpenEars due to hardware shortcomings. This allows optimizations that make OpenEars faster.
- ARC/LLVM compatible
- General recognition accuracy is improved, recognition speed is much faster in release versions due to LLVM
- Uses the latest .7 versions of all of the Sphinx libraries
- Calibration takes half the time and is configurable as an OpenEarsConfig.h constant
- Silence detection period is configurable programmatically
- It should no longer be necessary to do anything special in order to use iOS media objects such as AVPlayer.
- It is possible to set OpenEars to programmatically mix with background sounds such as iPod music, although this is not a default setting.
- OpenEars no longer attempts to start Pocketsphinx on a device that has no audio input available
- OpenEars now cleans dynamic language model arrays on input for problematic whitespace
- JSGF now has better perception
- Dynamic JSGF switching has been added
- The AudioSessionManager is now only used internally, since it was the main cause of configuration difficulty for most developers. OpenEars now manages the audio session automagically. Don’t call AudioSessionManager directly any more, it’s no longer a developer-facing class of OpenEars.
- All reported bugs fixed
- OpenEars has a license change and is no longer EFF/OSI-defined OSS. It is free as in beer so we’ll start calling it shared source instead of open source. The big license changes paraphrased (please read the new license since you are agreeing to it when you use the framework): you can’t redistribute it as source, using it in something other than an app requires permission and special licensing, contribution terms are spelled out. Although this affects a miniscule number of developers that have used OpenEars so far, it’s a change, so if absolutely anyone wants to talk this over with me/ask more about it, I sincerely welcome you to get in touch via the contact page and I will answer and discuss any good-faith question.
I really hope that you enjoy this easier, faster and more flexible 1.0 version of OpenEars!