Just want to confirm that 16 bit senone_score array (in structure acmod_s)
will always contain +ve numbers. Also 8 bit transition probabilities (in
structure tmat_t) will all be +ve numbers ?
My decoder is Pocketsphinx version 0.6.1
Thanks and regards,
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Just want to confirm that 16 bit senone_score array (in structure acmod_s)
will always contain +ve numbers.
Also 8 bit transition probabilities (in structure tmat_t) will all be +ve
numbers ?
It depends on the model (there could be very specific cases). For good
properly trained speech models they are positive.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
1) Going by Pocketsphinx code, senone_score is created by additions of
fast_logmath_add outputs (function
get_scores_4b_feat_4 for hub4wsj_sc_8k) and as per fast_logmath_add comments,
its output are +ve numbers and so
senone_score will always be a +ve number in this scenario ???
2) For the same model, Pocketsphinx initializes tmat probability scores, after
normalizing, flooring, log conversion
and quantization, the final values are assigned to uint8 variables so these
numbers will be +ve in this scenario ???
Regards,
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
Just want to confirm that 16 bit senone_score array (in structure acmod_s)
will always contain +ve numbers. Also 8 bit transition probabilities (in
structure tmat_t) will all be +ve numbers ?
My decoder is Pocketsphinx version 0.6.1
Thanks and regards,
It depends on the model (there could be very specific cases). For good
properly trained speech models they are positive.
Thanks NS,
1) Going by Pocketsphinx code, senone_score is created by additions of
fast_logmath_add outputs (function
get_scores_4b_feat_4 for hub4wsj_sc_8k) and as per fast_logmath_add comments,
its output are +ve numbers and so
senone_score will always be a +ve number in this scenario ???
2) For the same model, Pocketsphinx initializes tmat probability scores, after
normalizing, flooring, log conversion
and quantization, the final values are assigned to uint8 variables so these
numbers will be +ve in this scenario ???
Regards,