Hi, I'm starting a project porting Tremor to an ARM9 processor for an embedded system. I have a couple questions before I get too far along.

1. Are there any recommendations on test vectors for functional verification of the decoder? I have found a lot of references to vectors for comparative listening tests but nothing yet on vectors specifically designed to exercise all of the functionality of the decoder. I know that any sound file will work a lot of the algorithm but I also suspect that there are better source files for testing specific aspects of the decompression algorithm, as well as touching on corner cases.

2. Does anyone have experience with the difference between the standard and lowmem versions of Tremor? From what I've read it sounds like you lose about 10-15% data memory usage with lowmem with the cost of increased startup times. Does anyone have an actual value to attach to these latency differences? I know it's target dependent but any info is better than no info. Are there any other reasons to consider one branch or the other?

Thanks!

Ethan

(for those of you on the tremor list who might be seeing these questions twice, sorry!)

What i know is that the lowmem also has a lower quality, and usually is not required, as some of the mp3 players that support vorbis, are able to use the standard one.

Can't give any more details, as i don't have first-hand experience on it.

Are you sure you're talking about lowmem and not the _LOW_ACCURACY_ build switch? I thought that the sound quality of lowmem vs standard Vorbis was equivalent. _LOW_ACCURACY_ speeds up decoding by using 32bit rather than 64bit datatypes for intermediate values (I have a benchmark improvement of ~30% with _LOW_ACCURACY_ running on a PS2) but if the audio quality is unusable I won't be able to take advantage of that.

If you have other information to the contrary please let me know. I fully admit to being quite new at Vorbis, so if you believe something is misstated it probably is.

Errrr.....

Guess you're right.