Female recognition

[Andrew Porritt]

I'm developing an iPhone quiz game which allows players to answer
using VocalKit (as well as typing the answer). I've created a
dictionary for each question which only contains the words involved in
the answer, and this seems to work fine most of the time for male
voices, but it seems that female voices hardly work at all (we've
tried 10 male and 10 female, 8/10 males work fine, 1/10 females do).
I'm using the hmm folder from the VocalKit repository, and the
pocketsphinx.conf file (posted below), and dictionaries created via
the online lmtool software. Are there some settings in the
pocketsphinx.conf file that could help? I'm thinking maybe increasing
the sample rate or something. Has anyone experienced this before?
Thanks

<pocketsphinx.conf>

-fwdflat no
-bestpath no
-nfft 512
-lowerf 1
-upperf 4000
-samprate 8000
-nfilt 20
-transform dct
-round_filters no
-remove_dc yes

[Brian King]

I haven't done any testing, although that's a bit dissapointing. Did
they have accents as well? Sample rate probably won't have a huge
effect, and you need to update your accoustic models to match the
sample rate. You may want to try sniffing around on the voxforge.org
or the cmusphinx.org site and see if there are any other accoustic
models you can use. And it would be a great help if you report back
with what you find!

Brian King

> --
> You received this message because you are subscribed to the Google Groups "VocalKit" group.
> To post to this group, send email to voca...@googlegroups.com.
> To unsubscribe from this group, send email to vocalkit+u...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/vocalkit?hl=en.
>
>