Re: smallest page of encoded data

Thanks for the insight Carsten.
Basically, we're building a dictation system, with SPEEX at its core. Currently we have this working but the problem is, the SPEEX encoding process takes approximately 1 minute per MB of PCM audio, resulting in lengthy 'encode' jobs after a user has finished their dictation.
I am attempting to find a way around this, and my hope was that I could perform some type of incremental encode during the record process, yet still enable the user to trim/rearrange their dictation without requiring a complete re-encode.
And most users do edit the file to some degree...

Adam Langley
Senior Developer
Tel: +64 9 486 9010
[email protected]
www.winscribe.com
 
 

From: [email protected] [mailto:[email protected]] On Behalf Of Carsten Haese
Sent: Monday, 16 March 2009 2:08 p.m.
To: [email protected]
Subject: Re: [Vorbis] smallest page of encoded data

On Sun, Mar 15, 2009 at 8:27 PM, Adam Langley <[email protected]> wrote:
My end-goal is actually to do this with the SPEEX codec, but I'm using
Vorbis for now because I have previous experience with it. Am I going to
have the same problem with SPEEX?

You're going to have a different but equally insurmountable problem. Speex is a CELP-based codec. The LP stands for "Linear Prediction", which means that each frame depends on the history of frames that came before it. This in turn means that you can't rearrange frames without introducing artifacts.

By the way, you're not saying what kind of audio you're planning to encode, so it should be noted that Speex is purely designed to encode human speech. It's very good at that, but It will perform poorly if you try to encode music or other sounds with it.

-Carsten