I've been experimenting with trying to get phonemes to be properly detected. I've been doing this with several of my own audio files and had poor results. Then I tried with the provided goforward.raw file and it shows similar problematic results.
My install seems good, and it's working well for sentences:
% pocketsphinx_continuous -infile goforward.raw
go forward ten meters
But the -allphone option does not do what I expected.
% pocketsphinx_continuous -infile goforward.raw -allphone yes
SIL D SIL G OW F AO R W ER D JH T T EH N N M IY IH ZH ER Z S V SIL
It's not terrible, but there are some repeats and odd additions. Are there workarounds for this? Is this a common result? Do I need to tweak some options or the raw file?
I ultimately only wish to process a single word input, so any tips for accomplishing this are much appreciated.
System is Arch Linux with pocketsphinx 5prealpha. I've tried this with the source install and also the AUR package.
Use the command provided in documentation:
pocketsphinx_continuous -infile test/data/goforward.raw \
-allphone model/en-us/en-us-phone.lm.bin \
-beam 1e-20 -pbeam 1e-20 -lw 2.0