From: Robert H. <Rob...@gm...> - 2004-03-28 16:55:50
|
Hi Gabriel, Am Sonntag, 28. März 2004 15:39 schrieb Gabriel Bouvigne: > In calc_noise, we have this: > > > noise = *distort++ = noise / *l3_xmin++; noise on the right side is the quantization noise. noise on the left side is the relative noise compared to our target noise as our psy model will allow us. > /* multiplying here is adding in dB, but can overflow */ > noise = FAST_LOG10(Max(noise,1E-20)); > > It seems to me that it means that if noise==0 (on the first line), then > noise is becoming -20 because of the log10. > > I think that noise could be 0 in the following cases: > *quantized result == input data (a little unlikely) > *silent sfb > > In those 2 cases, we are considering the noise of the sfb to be -20, and > even worst we are adding -20 to total noise instead of 0. this is by intention, as there is room for adding even more noise in this band by increasing the quantization step size. > > > Am I wrong here? > > > Regards, Ciao Robert |