while iLBC is a narrowband speech codec, it utilizes the full 4 kHz frequency band available, whereas most standard low bit-rate codecs only use the frequency band from 300 Hz to 3400 Hz. This has a clear effect on speech quality.

--- cited from ILBC White Papers. GIPS.

About this,I have read it much time ago. And also I spoke it as else. But actually I did not get the meaning.

1. How and where do other codec,such as G7231.1 and G729, embody the idea "only use the frequency band from 300 Hz to 3400 Hz" ?

2. How and where does iLBC, embody the idea "utilizes the full 4 kHz frequency band available" ?
3. why are other codec,such as G7231.1 and G729, so silly to "only use the frequency band from 300 Hz to 3400 Hz",other than "utilizes the full 4 kHz frequency band available " ?

cheers
hyeewang@gmail.com

Please do not double post. Your duplicate post has been deleted.

IC. That is why I did not find the same post at speech codec section.

I don't know anything about these CODECs, but I would guess the bandlimiting is because these are lossy codecs. Some information will be lost... Something has to be thrown away...

Apparently the CODEC designers decided that the information in the 300-3400Hz band is more important (for intelligibility). If you throw-away all of the information outside that band, you can keep more information inside that band.

It also means that the CODEC doesn't have to "work" as hard (fewer CPU operations)... The CODEC doesn't have to decide moment-to-moment to keep or loose the information outside of the band. The decision has been made already.

This is crap from one end to the other. The codecs just encode what they get as input. The 3.4 kHz (more or less) low-pass is the anti-aliasing filter used for A/D or resampling. Sometimes the cut-off will be a bit higher, sometimes a bit lower. This has *nothing* to do with the codec. The 300 Hz filter (it's not a brickwall filter, more like a more or less gentle slope) is there because some codecs (including G.711) don't like to have very powerful bass. It's also useful in terms of quality because otherwise when you cut the high frequencies and keep all the low freq, the signal sounds strange (all shifted towards the low freq). So none of what I'm writing here has anything to do with iLBC, G729, or Speex for that matter. I'd even be surprised if the marketing paper really said that, I think you just extrapolated it. But if marketing indeed said that, I guess it's yet another proof you need to quit reading marketing doc (from any company) to get correct info.

Thank you jmvalin.

It is strictly cited from iLBC white paper, and I did not revise and extrapolate it.

I also have some questions about your remarks.

The 300 Hz filter (it's not a brickwall filter, more like a more or less gentle slope) is there because some codecs (including G.711) don't like to have very powerful bass. It's also useful in terms of quality because otherwise when you cut the high frequencies and keep all the low freq, the signal sounds strange (all shifted towards the low freq).

1. I knew some codec(G729),applying a highpass filter with cutoff frequency at 140 HZ to remove some undesired low frequency components. Here,you told G.711 don't like to have very powerful bass.
Why? Why it not handle and reconstrcut TRUE world speech signal?

2. "when you cut the high frequencies and keep all the low freq, the signal sounds strange (all shifted towards the low freq)."

-- please say it concretely. Why and how it sounds strange?

Thank you for your teaching.

1. The problem with G.711 is that it leaves a white coding noise with around 30 dB (or so) SNR. In most speech signal, the bass is often responsible for a significant fraction of the energy. Since the SNR is constant, the more energy, the higher the noise floor. So adding a lot of bass with G.711 create noise problems in the *high* frequencies (which are weaker and more susceptible to noise).

2. When you have very powerful bass and weak treble, the signal sometimes sounds even more "muffled" than if you cut the bass a bit. Just try it. I suspect it's depends a lot on the listener and probably other factors.