Vorbis PDA game audio

2003-09-20 13:10:08

Hey,

Here's the situation, there's this 'Lucasarts Virtual Machine' (opensource, http://www.scummvm.org) that allows you to play all kinds of old lucasarts adventures in pretty much every OS imaginable (dreamcasts, phones, linux, windows..) and is able to use ogg vorbis for audio compression. Also, scummvm is out for pocketPC.

Combine the two and you can compress the entire game (say) Day of the Tentacle to like 10Mb + audio data. Obviously you'd want to compress this audio (11khz, 16bit mono I think, mostly speech) the best possible way. And this results in a few questions:

1/ Should speex be used, even though some of the samples are non speech? ('sound effects')? I don't want to spoil my game experience by having fine speech but distorted by artefacts, and asking the programmers of scummvm to mix both will cost them considerable work - they have better things to do

2/ I'm interested in 'near-original' quality, not 'barely intelligible at extreme bitrates', for 11k, 16b, mono I think 32kbit should be fine, right? (speex or no)

3/ Right now, scummvm converts all audio samples from the big file into single files, then encodes each of the files into .ogg's, and then adds all those together again. This will result in fairly big codebook loss, won't it? One codebook for each file. Now one could conceivably first append all samples to one another, then encode that to one big ogg, and then seek inside that big .ogg. The question: won't we get strange effects with pre-echo and some parts of one sample 'spilling into' the other sample? And if so, any way to prevent this?

Thanks