Home › Forums › OpenEars › Increasing speech recognition accuracy using a custom languagemodel? › Reply To: Increasing speech recognition accuracy using a custom languagemodel?
Welcome,
Do you think it would help to make a custom languagemodel using the lmtool?
No, there is no accuracy difference between OpenEars custom models and the models from that tool, which are both generated using CMU language model generation. It would probably be more effective to share your goals and issues here and see if I can help you.
Regarding your paths being null, this just needs the normal kind of troubleshooting you’d need to do to make sure that any app bundle file path had a real file at the end of it – check that the items are really added to the target and that you’re running the target you think you are, find out where to expect an added file to be found in the bundle and how that is referenced (in this case, you look like you are trying to get a file from a caches directory, but normally a file that you add to an app target is found in the app bundle rather than the sandbox user files so I would probably start there with investigating the issue, plus you must already know the full names of your files since you named them before adding them to the bundle so that would be the second thing to look into, since your name appears to be dynamically generated), and use NSFileManager to verify the existence of your file or return an error.
OpenEars’ file loading is known to work fine using normal NSString representations of file paths, so you can just focus on figuring out why you have a null path, which is going to relate to app bundle structures/Cocoa APIs rather than OpenEars.