Arabic Language Model - WORD ERROR RATE: 89.1%

[ehab ahmad]

hello every one
i've been able to build an arabic LM after reading the tutorial here
with about 5 different voices and about 2 minutes of recordings
the language model aims to recognize arabic digits but i have a very low accuracy and only the last word is being detected?
what i should do to improve the accuracy ?
please help

[Nickolay V. Shmyrev]

Add more training data, not necessary digits.

[ehab ahmad]

Hello : Nickolay V. Shmyrev
Update: Please help "it's been 2 months trying to get a result :)"
i have a 202 word now and 34 sentences and 34 recordings
the LM is built but training faild with this error ON ALL MY RECORDINGS :

Failed to align audio to trancript: final state of the search is not reached

this is the whole project in the attachments and the log file is included
what am i doing wrong???

[Nickolay V. Shmyrev]

You do not have enough data for training. Tutorial recommends at least 10 hours of speech for single speaker.

[ehab ahmad]

on every recording i add i got the same error that the recording is ignored and it faild to align with the transcription
shouldn't i get a high WER warning instead of this errors if the recordnigs number was the cause of this and they werent enough??

[Nickolay V. Shmyrev]

shouldn't i get a high WER warning instead of this errors if the recordnigs number was the cause of this and they werent enough??

No