Sir but it shows error for missing mixture_weights file.
Sir I didn't get idea. Excuse me sir. Once again explain me sir.
My voice indian accent based. It gives 50-55% accuracy only for correct
inference.
So I want to train my own digits audio. It will adopt or not? How can I get
this mixture_weights file.?
Which way to achieve accuracy? Sir please help me sir.thank you so much.
Thanks a lot sir.
Last edit: Nickolay V. Shmyrev 2019-02-19
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
You didn't answer my question. You wrote you want to try the training but you are doing adaptation. Figure out what you need to do first and we'll gladly help you.
Last edit: Nickolay V. Shmyrev 2019-02-19
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
first of all really sorry sir. i didn't understand your first reply. so i am not answering to you properly.
You wrote you want to try the training but you are doing adaptation.
yes sir. retrain to adapt my audio features for pocketsphinx tidigits AM,
Figure out what you need to do first and we'll gladly help you.
sir i want to adapt my indian accent acoustic sound features with tidigits pocketsphinx acoustic model.
reason behind,
1) i don't have much dataset like indian accent based digit audios.
sir but how we will partularly adapt default english model to digit.
My idea is:
i already did this,
i was adapted with own dataset with default en-us & en-in models. but that not gives good accuracy. that en-us-dict not particular digits based and lm also,. it gives irrelavent words also.
and i tried to build own LM & Dict using CMU-toolkit.
it is also failed for me.
any possibities is there or not sir, we should adapt minimum dataset with digits LM& Dict.
if we have to train a new acoustic model for tidigit, then how much hours vs speakers audio datasets must.
thank you very much sir.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
how can i trained tidigits acoustic model with my own audiofiles?
missing mixture_weights file in the acoustic tidigits model.
i was tried to train my own accent audios for digits using cmu pocketsphinx.
i was struggled lot, in this part. no such file mixture_weights
sphinx_fe -argfile tidigits/hmm/feat.params -samprate 16000 -c digits.fileids -di . -do . -ei wav -eo mfc -mswav yes pocketsphinx_mdef_convert -text tidigits/hmm/mdef tidigits/hmm/mdef.txt cp -a /usr/local/libexec/sphinxtrain/bw . cp -a /usr/local/libexec/sphinxtrain/mk_s2sendump . cp -a /usr/local/libexec/sphinxtrain/map_adapt . cp -a /usr/local/libexec/sphinxtrain/mllr_solve . ./bw \ -hmmdir tidigits/hmm \ -moddeffn tidigits/hmm/mdef.txt \ -ts2cbfn .cont. \ -feat 1s_c_d_dd \ -cmn current \ -agc none \ -dictfn tidigits.dic \ -ctlfn digits.fileids \ -lsnfn digits.transcription \ -accumdir . ./mllr_solve \ -meanfn tidigits/hmm/means \ -varfn tidigits/hmm/variances \ -outmllrfn mllr_matrix -accumdir . mkdir en_in_own cp -a tidigits/hmm/ en_in_ownhow to i resolve this issue. and then how to train with my own accent besed digits audios.
help me sir. i am begginer for this concept.
Thanks.
From the commands it seems you are trying to adapt the model instead of training it.
Sir but it shows error for missing mixture_weights file.
Sir I didn't get idea. Excuse me sir. Once again explain me sir.
My voice indian accent based. It gives 50-55% accuracy only for correct
inference.
So I want to train my own digits audio. It will adopt or not? How can I get
this mixture_weights file.?
Which way to achieve accuracy? Sir please help me sir.thank you so much.
Thanks a lot sir.
Last edit: Nickolay V. Shmyrev 2019-02-19
You didn't answer my question. You wrote you want to try the training but you are doing adaptation. Figure out what you need to do first and we'll gladly help you.
Last edit: Nickolay V. Shmyrev 2019-02-19
first of all really sorry sir. i didn't understand your first reply. so i am not answering to you properly.
yes sir. retrain to adapt my audio features for pocketsphinx tidigits AM,
sir i want to adapt my indian accent acoustic sound features with tidigits pocketsphinx acoustic model.
reason behind,
1) i don't have much dataset like indian accent based digit audios.
Actually i was referred for CMU -Sphinx
https://cmusphinx.github.io/wiki/tutorialadapt/
own way to tried this,
how do i adapt model with own audiofiles. Thanks a lot sir for your quick response.
Thanks.
Last edit: Murugan R 2019-02-19
Adapt the default english model to digits, it will be ok.
sir but how we will partularly adapt default english model to digit.
My idea is:
i already did this,
i was adapted with own dataset with default en-us & en-in models. but that not gives good accuracy. that en-us-dict not particular digits based and lm also,. it gives irrelavent words also.
and i tried to build own LM & Dict using CMU-toolkit.
it is also failed for me.
any possibities is there or not sir, we should adapt minimum dataset with digits LM& Dict.
if we have to train a new acoustic model for tidigit, then how much hours vs speakers audio datasets must.
thank you very much sir.
Same as in tidigits dataset
sir that digit phonesets is from one accent to other accent is same or different?
like en_us vs en-in acoustic phonesets.
Thanks sir.
Last edit: Murugan R 2019-02-19
It could be the same.
Mm k sir. Thank you very much sir. I will update my status if I am getting
good accuracy for digits based recognition in indian accent.
Thanks.
On Tue 19 Feb, 2019, 8:31 PM Nickolay V. Shmyrev <
nshmyrev@users.sourceforge.net wrote: