**** BEGIN LOGGING AT Thu May 10 03:00:02 2018 May 10 13:41:24 Ok the Bela folks are on a roll here - https://blog.bela.io/2018/05/08/CTAG-multichannel-audio-cape/ May 10 13:41:29 8 in 16 outs! May 10 13:52:43 ds2 I recorded my speech of 21 words...and tried to find out the accuracy....using pocketsphinx_batch and then word_align...the results were horrible...none of the words have been recognized...I saw the hypothesis file...it is also true that I am using just my earphone mic to record my audio...I think the results will significantly improve if I use AGC mic May 10 17:17:43 anirban1998[m]: 0/21 is pretty bad... have you tried looking at the speech with a tool like audacity? see if it is really low, distorted, etc? May 10 17:19:29 ds2 Will check with audacity and inform May 10 17:20:45 ds2 Actually I think the problem is with the quality of my mic...sometimes the words are recognized...sometimes not May 10 17:21:39 anirban1998[m]: I don't disagree, but looking at it with a tool should give more insight - the mic can be distorting or just too low in signal level May 10 17:22:29 it may be interesting (if you have time) to normalize the recording in audacity then run that wave file through sphinx May 10 17:22:49 Ok May 10 17:23:13 what sample rate are you using? 16000? May 10 18:51:34 No I am using 41000...will there be any problem? May 10 18:54:45 ds2 If there is any problem due to the sampling rate then I will again check out the accuracy with 16000 sampling rate May 10 22:44:12 jkridner[m]: anujdeshpande[m] : https://github.com/jadonk/bonescript/pull/165 , can you please look at this. May 10 22:45:33 I have replaced the use of exports in the library with module.exports , as suggested in jadonk/issue#82 May 10 23:26:49 anirban1998[m]: it is a matter of noise... the faster the sample rate, the wider the bandwidth (see Nyquist, et al.). I wonder if a software filter with a corner freq near 3.8KHz (number from phone system) would help **** ENDING LOGGING AT Fri May 11 03:00:03 2018