I'm testing the new trie LM vs. the old DMP and on some classroom lectures I'm seeing an absolute WER improvement of about 3-10% depending on topic, so thank you very much for this update!
I was wondering though:
1) Is it possible to get a version of this model with no pruning or less pruning?
2) On another topic, is there a continuous 8kHz version of the cmusphinx 5.2 acoustic model? The downloads page only has a PTM version of the 5.2 8kHz model and a continuous version of the old 5.1 model. An improved continuous 8kHz 5.2 model might help squeeze out a few more % WER improvement.
Thanks again.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm testing the new trie LM vs. the old DMP and on some classroom lectures I'm seeing an absolute WER improvement of about 3-10% depending on topic, so thank you very much for this update!
Great, you are welcome
1) Is it possible to get a version of this model with no pruning or less pruning?
It is available on our torrents (US English Language model)
2) On another topic, is there a continuous 8kHz version of the cmusphinx 5.2 acoustic model? The downloads page only has a PTM version of the 5.2 8kHz model and a continuous version of the old 5.1 model. An improved continuous 8kHz 5.2 model might help squeeze out a few more % WER improvement.
Sure, I can train it in a week or so, it will just take some time.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi all,
I'm testing the new trie LM vs. the old DMP and on some classroom lectures I'm seeing an absolute WER improvement of about 3-10% depending on topic, so thank you very much for this update!
I was wondering though:
1) Is it possible to get a version of this model with no pruning or less pruning?
2) On another topic, is there a continuous 8kHz version of the cmusphinx 5.2 acoustic model? The downloads page only has a PTM version of the 5.2 8kHz model and a continuous version of the old 5.1 model. An improved continuous 8kHz 5.2 model might help squeeze out a few more % WER improvement.
Thanks again.
Great, you are welcome
It is available on our torrents (US English Language model)
http://cmusphinx.info
see also
http://cmusphinx.sourceforge.net/2013/01/a-new-english-language-model-release/
Sure, I can train it in a week or so, it will just take some time.
Thanks again, I'm looking forward to trying it.
Hi, Is the new trie LM available for windows version of sphinxbase? The last modified date for the win32 zip is showing feb 2015.
You can compile from sources.
Thanks Nickolay. I will do that. Any idea when will the windows version be released?