Poor recognition accuracy of Pocketsphinx using phoneme recognition on Android, French language -

i working on project have integrate speech functionalities of pocketsphinx android application. in fact, have integrate phoneme recognition functionality provided pocketpshinx should able recongize phonemes in french language, e.g. speech recongnizer should able recognize syllabes( "de", "re", "se", etc), consonants (like "m", "f", "g", etc), double-consonants(like "kl", "ks", "gr",etc) , vowels(like "a", "o","e",etc). right now, have integrated pocketsphinx recognizing phonemes mentioned above, have bad results. example, when pronounce "o", recognized result is: "sil ff ei au" (even did not pronounce letter "f" , "e" @ all), or else appeared @ beginning not pronounced. letters appeared @ beginning not same (sometimes "ll", "uu", etc). sometimes, letter pronounce appears @ beginning (e.g. "a", "sil aa sil") , happens rarely. so, guys please , let me know problem , suggestions solving problem? need create language model , dictionary these letters interested in (since number of letters fixed)? or, maybe need adaption of acoustic model, or train acoustic model?

thank in advance!

leutrim

in model find feat.params , change -cmuinit value according choice, also, value of "-kws_threshold" 1e-40f removing false alarms.