Closed AndyZe closed 7 years ago
I would like it if you create a "kinetic" branch for the newest ROS release. Then I would do a PR for that branch, too.
Can I also ask, are the predictions very accurate for you? They are not accurate at all for me -- most of the predictions are "half speed." Maybe it's been trained for you, though.
Well, you are right. The predictions do not work correctly most of the times. In my case, if I speak at the right moments and with proper phonetics, the system is able to detect "right", "left", "half speed" much more accurately. Currently, this repo uses the same recogniser as the original one. I am currently working on improving the accuracy. But it might take some time before it is completed.
Thanks for reviewing. You've done the community a pretty good service by releasing a ROS Kinetic-compatible version, even if that wasn't intentional. ;)
I'm getting much better accuracy after changing the audio channel format. Read here: http://cmusphinx.sourceforge.net/wiki/tutorialtuning
This is the change I made:
# See http://cmusphinx.sourceforge.net/wiki/tutorialtuning
stream = pyaudio.PyAudio().open(format=pyaudio.paInt16, channels=1,
rate=16000, input=True, frames_per_buffer=65536)
I'll make another pull request, if you want (I would prefer to have a Kinetic branch, though, because that's what I'm using)
Thanks for the feedback. And the article. I will read it soon. I have my semester exams starting next week because of which I haven't been able to actively contribute here. I will try to update this repo with the Kinetic branch, possibly this weekend.
minor stuff...