I didn't try with English english because I just downloaded American english... Where can I find other languages? (I'm French so it would be better if I had the french language model).
I create my sound files with this command:
arecord -r -f s16l -s 16000 sound_file_name.16k
Is it correct?
Thanks.
julie
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2002-04-05
Sorry, have no idea were you might find French lm.
As for the sound file format. Run gofoward.16k through aplay, as you seem to have alsa install. It should tell you what the recording options were. Then run you sound files through and see that they have the same options.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2002-04-18
I also generated my own corpus of 1000 words using QuickLM.
Downloaded hub4 LM from CMU web site and pushed my pre-recorded sound files at it using shphinx-test. The recognition rate is terrible. 1 in 10 sentences recognised.
Anybody any ideas on how to improve the quality of recognition
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello,
I created my own corpus (500 words of vocabulary) and generated the .lm and .dic files thanks to Quick LM web site.
I run sphinx2-test (from version 0.4) with my own sound files and sphinx recognizes 50% sentences only.... :-(
So, I tried with the former example (turtle) and its own corpus and LM files... but the quality is not better.
Has anybody good performance with sphinx2-0.4 ? In this case, could you explain me how to improve mines?
Thanks a lot
JD
Does it depend on the sound files you are using ? Would it recognise American english better than English english, or doesnt it matter.
I didn't try with English english because I just downloaded American english... Where can I find other languages? (I'm French so it would be better if I had the french language model).
I create my sound files with this command:
arecord -r -f s16l -s 16000 sound_file_name.16k
Is it correct?
Thanks.
julie
Sorry, have no idea were you might find French lm.
As for the sound file format. Run gofoward.16k through aplay, as you seem to have alsa install. It should tell you what the recording options were. Then run you sound files through and see that they have the same options.
I also generated my own corpus of 1000 words using QuickLM.
Downloaded hub4 LM from CMU web site and pushed my pre-recorded sound files at it using shphinx-test. The recognition rate is terrible. 1 in 10 sentences recognised.
Anybody any ideas on how to improve the quality of recognition