On the heels of mp3pro's respectable performance in the 64 kbit/s tests, I was hoping someone could give meus a better understanding of what the SBR process entails and what info is actually carried regarding high frequencies. I'd like to get beyond the press release hype and basics from Coding Technologies and Thomson.

I think I get the basics(but I am certainly open to correction on this): Per CT, in mp3pro, SBR acts like a shell around the mp3 codec. So, in pre-encoding stage, a bitrate specific threshold is determined( seems to be ~ 8.1 kHz at 64 kbits/s and ~10 kHz at VBR 100%/Highest Quality (which seems to average ~130-140 kbits/s), and frequencies above will be sampled at 44.1 kHz and then filtered out after some info is stored in the ancillary bits of the mp3 stream, Then, frequencies below this threshold will be encoded as mp3 sampled at 22kHz. After mp3 decoding, the SBR information is applied to the decoded midrange frequencies to get a new high frequency spectrum that is based on harmonics of the existing frequencies.

So I'm left with questions:

What information can you put into a few ancillary bits(even recognizing that SBR is largely a post-decoding process?

Aren't there fundamental frequencies >10 kHz as opposed to harmonics that are filtered and not replicated?

How can this process deal with overtones that aren't harmonics?

How accurate can the relative amplitude of these replicated frequencies be?

What other obvious things am I missing?

I'm not slamming mp3pro nor am I heaping praise on it. I was surprised at how well it handled some samples and how poorly it handled others in the 64 kb/s test. Think I saw rjamorim comment to the effect that this may be a poor implementation of good technology. Certainly mp3pro has problems aside from SBR that rest with the underlying FhG codec:)

I'm a little bummed that I didn't get any responses on this since there seemed to be a little discussion about SBR after the 64 kb/s test results.... Can anyone add a few more details about SBR, please ?

SBR - Spectral Band Replication

Guessing the nontransmitted higher frequency range of a compressed audio file by using some helper bits (transmited with the stream) in addition to the transmitted base band. SBR allows a restoration (not reconstruction) of the upper frequency range without lots of bits. It was developed by Coding Technologies, and is useful for medium and high quality coding (although the latter is arguable) at low and medium data rates. It is used by mp3PRO and Digital Radio Mondiale.

SBR uses the fact that most instruments in music have harmonics, which means that a strong signal at frequency f on the spectrum is likely to have 'echos' at frequencies 2f, 4f, 8f, etc. Therefore SBR works well for instrumentals, but not so well for samples which have lots of non-harmonic sounds like percussion, voices, etc.