Is there a suggested acoustic model to use for embedded/mobile devices?
I'm using pocketsphinx 0.5 (release, not SVN) on an ARM11 (iPhone) system. It takes roughly 10 seconds to load the wsj_sc8k model (mdef init) and about 2 minutes to load cmudict 0.6d. Using a smaller dictionary definitely reduces dict_load and ngram_search_update... times, but the model definition init is constant.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks for the advice, but training a model is a bit more involved than I would like to take this step (and perhaps would lead to less accuracy). After digging a bit more, I realized the model definition file is supported in binary and pocketsphinx provides pocketsphinx_mdef_convert to convert between binary and text mdef files. The execution time was drastically improved (as expected) when I switched to a binary mdef file. I'll check out the SVN sources to see if binary dictionary file is supported. If not, I'll write it myself.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Is there a suggested acoustic model to use for embedded/mobile devices?
I'm using pocketsphinx 0.5 (release, not SVN) on an ARM11 (iPhone) system. It takes roughly 10 seconds to load the wsj_sc8k model (mdef init) and about 2 minutes to load cmudict 0.6d. Using a smaller dictionary definitely reduces dict_load and ngram_search_update... times, but the model definition init is constant.
You can train your own one from voxforge data. Smaller number of senones and gaussians will speedup things.
Thanks for the advice, but training a model is a bit more involved than I would like to take this step (and perhaps would lead to less accuracy). After digging a bit more, I realized the model definition file is supported in binary and pocketsphinx provides pocketsphinx_mdef_convert to convert between binary and text mdef files. The execution time was drastically improved (as expected) when I switched to a binary mdef file. I'll check out the SVN sources to see if binary dictionary file is supported. If not, I'll write it myself.
Hi to all,
Can any one tell me the procedure to run pocket sphinx in wince mobile ??
i already worked in SPHINX-4 for indian accent english and achieved that......
please tell me the step by step procedure so that i can run pocket sphinx.
as we are doing research on pocket sphinx,by sharing our knowledge together we will achieve our goals
thank you inadvance.....
Please stop spamming our forums.