>
> Here is a proposal to help choosing the quantization.
>
> We can use the audition threshold. For each quantization, add the minimal
> audition thresholds corresponding to the frequency of each distortion. For a
> given over value, the quantization with the lowest sum would be the best.
>
> For the value of this threshold, I suggest using a table giving the minimum
> of the audition threshold in the frequency range of each subband. The
> following formula gives the human minimal audition threshold of a young
> people with very good hearing, f beeing the frequency:
>
> 3.64*(f/1000)^(-0.8)-6.5*e^(-0.6)*(f/1000-3.3)^2+10^(-3)*(f/1000)^4
>
By adution threshold, I assume you mean what I call the threshold in
quiet? This would actually be very easy to implement. I modified
calc_noise a while ago to return the total quantization noise in each
band. With that formula (do you have a reference?) it would be easy
to compute the noise above the adution threshold and calculate a new
quality metric to use when comparing quantizations with the same value
of "over".
I think I will also use that formula for -f mode:
Instead of using masking thresholds of 0, we could just use the
adution threshold.
>
> However, I can't compute the table values, as I don't know the frequency
> range of each subband, and I didn't find it in the iso doc.
>
You have to do some work to decifer it: look up the MDCT coefficients
which belong in each band. Knowing the sampling rate and MDCIT coefficient
index you can compute the frequency.
Mark
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
