Speech Research, A quick question on the AMR codec

Hi there...

I am undertaking some research into the AMR codec in background noise. I
have found that the codec seems to suppress background noise as I am
seeing a lower SNR and a audible increase in quality.

However, I am using the C-reference code (3GPP TS 26.073) and it is my
understanding that this reference implementation does NO noise
suppression. Noise suppression is an optional unit defined in some other
3GPP TS.

So therefore, I am concluding that the CELP style AMR codec has some
type of implicit noise suppression, due to the nature of the CELP
coding. I suppose this can be explained by the fact that the codebook
used in CELP is trained on speech, and therefore, noisy speech will get
'mapped' to the closest speech like sound.

My question is, does this make sense? Has anyone else out there found a
similar observation?

I ask, because it kind of goes against all that I would have expected.

Thanks

Alastair James

Sort of. The assumption of "line spectral pairs" means that anything
with more than two harmonic components gets limited to the loudest two
such formants and their harmonics. Since voiced sounds make up the
majority of speech, this causes a form of noise suppression.

To get an idea of the disadvantage, compare how two people talking at
the same time sound using CELP and PCM (in that order; if you listen
after you know how they are supposed to sound, the difference will
seem to be less distinct -- have someone else clip a recording of two
simultaneous voices and CELP-codec it before you listen.)

Sincerely,
James Salsman
--
www.readsay.com - maker of the ReadSay PROnounce English literacy system
400 MHz PDA included: $499 -- http://www.readsay.com/PROnounce.html

Thanks for that!

Alastair