In pocketsphinx, CMN_WIN_HWM and CMN_WIN have been defined as 800 and 500
respectively. I have observed that there is no change in accuracy even if the
values are defined as low as 8 and 5 respectively. Infact with the limited
dataset which I have, I didnot find any performance issue even with values of
1 and 0 with pocketsphinx_batch. Although the performance of
pocketsphinx_continuous was very poor with low values of 0 and 1. In
continuous mode the performance becomes quite OK at values of 8 and 5 and
above for CMN_WIN_HWM and CMN_WIN.
Is there any tradeoff between higher values and lower values (other than
memory requirements).
Regards
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Yes, this change could work for you in your particular application but I
suggest you to leave this part as is. I think that the memory optimization
from this change is not significant, it's basically not worth doing that.
If you want to optimize memory, you could probably check where the biggest
chunk is allocated with massif or some other memory profiling tool.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
In pocketsphinx, CMN_WIN_HWM and CMN_WIN have been defined as 800 and 500
respectively. I have observed that there is no change in accuracy even if the
values are defined as low as 8 and 5 respectively. Infact with the limited
dataset which I have, I didnot find any performance issue even with values of
1 and 0 with pocketsphinx_batch. Although the performance of
pocketsphinx_continuous was very poor with low values of 0 and 1. In
continuous mode the performance becomes quite OK at values of 8 and 5 and
above for CMN_WIN_HWM and CMN_WIN.
Is there any tradeoff between higher values and lower values (other than
memory requirements).
Regards
Hello
Yes, this change could work for you in your particular application but I
suggest you to leave this part as is. I think that the memory optimization
from this change is not significant, it's basically not worth doing that.
If you want to optimize memory, you could probably check where the biggest
chunk is allocated with massif or some other memory profiling tool.