Speex dev - Jun 2006 - seeking in time, feedback+question

Hi all,
Since this is my first e-mail to the list, let me
start with: Thanks for this wonderful project!!! very
useful indeed.

I was trying to compress some big files and then to
access to small portions and decode only those parts.
I use the streaming mode(the sample codes provided on
the web-site) without the Ogg packaging which seems
useful to me for seeking in time.
When I seek in time by just skipping packages, reset
by: 
speex_decoder_ctl(state, SPEEX_RESET_STATE, NULL);
and starting decoding, I see that the quality of
decoding varies depending on the initial frame
characteristics. It seems that the signal gets close
to the original quite fast when the first frames are
unvoiced. If I start decoding in the middle of a
voiced chunk then decoding quality is comperatively
much lower, more frames are needed until the signal
gets closer to the original. 

This was the feedback part. So the obvious question
is:
do you know an easy way to get rid of this problem?
Thanks.
Regards,
Baris

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com

> I was trying to compress some big files and then to
> access to small portions and decode only those parts.
> I use the streaming mode(the sample codes provided on
> the web-site) without the Ogg packaging which seems
> useful to me for seeking in time.
> When I seek in time by just skipping packages, reset
> by:
> speex_decoder_ctl(state, SPEEX_RESET_STATE, NULL);
> and starting decoding, I see that the quality of
> decoding varies depending on the initial frame
> characteristics. It seems that the signal gets close
> to the original quite fast when the first frames are
> unvoiced. If I start decoding in the middle of a
> voiced chunk then decoding quality is comperatively
> much lower, more frames are needed until the signal
> gets closer to the original.
>
> This was the feedback part. So the obvious question
> is:
> do you know an easy way to get rid of this problem?
I do not know how you could get rid of the problem, but you could mask it by 
starting the decode a few frames early, and throwing away the audio for 
those early frames.  Then the decode state will have converged when you 
start decoding the audio that you want.

 Regards,

Jim Crichton