Hello!
I am encoding small snippets of audio (e.g. 100ms) that
contain important audio from the first sample till the last one.
(means they dont start/end silent).
Doing so raised a couple of questions I couldnt solve
by reading the docu/faq/internet search.
1. Does the "complexity" parameter influence only the
speed of the encoder or also the speed of the decoder?
(I need fast decoding, but have a lot of time for encoding).
2. Does Speex use information from the previous frame to encode the
next frame, or are they totally independent?
I know that I can decode the frames independently, but what about the
encoder?
3. I noticed that when I encode+decode, SPEEX inserts about
~16ms of silence at the beginning, this is probably the lookahead time.
I am chopping this manually off right now, but the codec should
do this transparently.
First because your audio data is time shifted, and second
you need to store more audio data than the original audio has,
to not clip the end.
There is another effect: because the lookahead is not equal to the
framesize (looks like about half of it), the actual audio you feed in
starts in the middle of the first frame, were the first half is zero
padded.
That causes Speex to somehow "fade in" the audio, since it assumes
comming from silence.
The only ad-hoc solution that came to my mind is to feed in
#(framesize-lookahead)
samples first, to generate the first "garbadge" frame, throw it away
and
proceed as usual.
This makes the usage of Speex pretty clumsy, if your constraints are
"what you feed is what you get".
Any ideas, or am I doing something wrong?
Regards,
Thilo Koehler
--
