 | | From: | amosgcs | | Subject: | speaker verification library | | Date: | Wed, 19 Jan 2005 08:52:44 +0200 |
|
|
 | hi,
I have a library for speaker verification which works fine with human
voices. I know that it uses Cepstrum and vector quatization modeling for the
calculations. My question:
Does anyone knows whether its usefull also with synthesized voices like
bell, siren, rings?
Thanks in advance
Amosgcs
|
|
| | From: | Sathish Athreya | | Subject: | Re: speaker verification library | | Date: | Wed, 19 Jan 2005 10:41:41 +0100 |
|
|
| May not be.
"amosgcs" wrote in message
news:csl06v$so5$1@news2.netvision.net.il...
> hi,
>
> I have a library for speaker verification which works fine with human
> voices. I know that it uses Cepstrum and vector quatization modeling for
the
> calculations. My question:
> Does anyone knows whether its usefull also with synthesized voices like
> bell, siren, rings?
>
> Thanks in advance
>
> Amosgcs
>
>
>
>
|
|
| | From: | John Openshaw | | Subject: | Re: speaker verification library | | Date: | Thu, 20 Jan 2005 23:06:22 +0000 |
|
|
| In article , amosgcs
writes
>hi,
>
>I have a library for speaker verification which works fine with human
>voices. I know that it uses Cepstrum and vector quatization modeling for the
>calculations. My question:
>Does anyone knows whether its usefull also with synthesized voices like
>bell, siren, rings?
It depends. The cepstrum is meant (in theory) to model the vocal tract
as a filter and attenuate the excitation that produces speech through
it. So if your signals are very impulsive then the cepstrum you use may
throw out a lot of discriminatory information. If your system is capable
of being configured so that it uses the higher order cepstra that might
help.
But at the end of the day just give it a go.
Cheers
--
John Openshaw
|
|
| | From: | Tomi Kinnunen | | Subject: | Re: speaker verification library | | Date: | Wed, 19 Jan 2005 19:47:48 +0000 (UTC) |
|
|
| Why not, go and try it out!
Cepstrum models the shape of the spectrum so I assume it would
work also for other sound sources as they have some characteristic
spectral shapes.
-Tomi
: hi,
:
: I have a library for speaker verification which works fine with human
: voices. I know that it uses Cepstrum and vector quatization modeling for the
: calculations. My question:
: Does anyone knows whether its usefull also with synthesized voices like
: bell, siren, rings?
:
: Thanks in advance
:
: Amosgcs
:
:
:
:
--
|
|
