Regarding the Improved CMN I've started to work on it since it's highly related to our work at the company and I'm going to release a patch for it once I have a good working solution.
If somebody already has a patch it would be even better.
I've scratched few notes regarding the problem you can see the attachments if you're curious. I will try to solve it in a generic way that hopefully helps everybody.
Nice start, however, your assumptions about CMN are not quite correct. CMN is a generic property of a channel, it has no relation to the recorded device. Even if user changes the recording level you need to adjust CMN accordingly. For example if user changes the distance to the microphone.
You should probably need to review CMN theory and extensions first. You can get some useful references from Chanwoo Kim thesis on PNCC for example.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Nice start, however, your assumptions about CMN are not quite correct. CMN is a generic property of a channel, it has no relation to the recorded device. Even if user changes the recording level you need to adjust CMN accordingly. For example if user changes the distance to the microphone.
Very informative. I wasn't actually sure about my assumption so this greatly changes my methods.
I will review Chanwoo Kim thesis on PNCC but most likely I will go to your suggestion to brute-force CMN estimation. It fits easier in the brain of an application developer from the deterministic world, besides it's a black-box approach.
Last edit: Omar Al-Ithawi 2013-12-08
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Regarding the Improved CMN I've started to work on it since it's highly related to our work at the company and I'm going to release a patch for it once I have a good working solution.
If somebody already has a patch it would be even better.
I've scratched few notes regarding the problem you can see the attachments if you're curious. I will try to solve it in a generic way that hopefully helps everybody.
You feedback is appreciated.
Last edit: Omar Al-Ithawi 2013-12-05
Pls test the link , wether it is working or not
Last edit: Nickolay V. Shmyrev 2013-12-05
Nice start, however, your assumptions about CMN are not quite correct. CMN is a generic property of a channel, it has no relation to the recorded device. Even if user changes the recording level you need to adjust CMN accordingly. For example if user changes the distance to the microphone.
You should probably need to review CMN theory and extensions first. You can get some useful references from Chanwoo Kim thesis on PNCC for example.
Very informative. I wasn't actually sure about my assumption so this greatly changes my methods.
I will review Chanwoo Kim thesis on PNCC but most likely I will go to your suggestion to brute-force CMN estimation. It fits easier in the brain of an application developer from the deterministic world, besides it's a black-box approach.
Last edit: Omar Al-Ithawi 2013-12-08
@nicolay I didn't do any progress on this task. We ended up ditching and use sphinx as-is.