It seems silly to regenerate this file every time Kaylee starts. If the voice commands haven’t changed, the file will be exactly the same as before, but we created extra I/O. Sure the file won’t be huge, so it’s not a big deal, but it’s a silly inefficiency when we’re already hashing files to see if things have changed.
A possible solution would be to generate the file in memory, and if the hash differs from the stored one, write the generated file out to disk. This would allow us to do things like remove duplicate words, though the lmtool doesn’t seem to care if we have duplicate words so why should I?
It seems silly to regenerate this file every time Kaylee starts. If the voice commands haven't changed, the file will be exactly the same as before, but we created extra I/O. Sure the file won't be huge, so it's not a big deal, but it's a silly inefficiency when we're already hashing files to see if things have changed.
A possible solution would be to generate the file in memory, and if the hash differs from the stored one, write the generated file out to disk. This would allow us to do things like remove duplicate words, though the lmtool doesn't seem to care if we have duplicate words so why should I?
Note: According to this CMUSphinx Wiki article, we really shouldn’t remove duplicate words from the corpus, and in fact we should do something smarter about numbers than we’re doing now. That should be a separate issue, though.
Note: According to [this CMUSphinx Wiki article](http://cmusphinx.sourceforge.net/wiki/tutoriallm), we really shouldn't remove duplicate words from the corpus, and in fact we should do something smarter about numbers than we're doing now. That should be a separate issue, though.
clara
It seems silly to regenerate this file every time Kaylee starts. If the voice commands haven’t changed, the file will be exactly the same as before, but we created extra I/O. Sure the file won’t be huge, so it’s not a big deal, but it’s a silly inefficiency when we’re already hashing files to see if things have changed.
A possible solution would be to generate the file in memory, and if the hash differs from the stored one, write the generated file out to disk. This would allow us to do things like remove duplicate words, though the lmtool doesn’t seem to care if we have duplicate words so why should I?
Note: According to this CMUSphinx Wiki article, we really shouldn’t remove duplicate words from the corpus, and in fact we should do something smarter about numbers than we’re doing now. That should be a separate issue, though.