[FFmpeg-devel] [PATCH 2/3] [GSoC] [AAC] aaccoder: Implement Perceptual Noise Substitution

[Claudio Freire]

On Sun, Apr 12, 2015 at 1:50 AM, Rostislav Pehlivanov
<atomnuker at gmail.com> wrote:
> @@ -245,6 +258,8 @@ static float (*const quantize_and_encode_band_cost_arr[])(
>      quantize_and_encode_band_cost_UPAIR,
>      quantize_and_encode_band_cost_UPAIR,
>      quantize_and_encode_band_cost_ESC,
> +    NULL,
> +    quantize_and_encode_band_cost_NOISE,
>  };


Make a function that has the same signature as the
quantize_and_encode_band_cost_X functions here, but whose only
contents are an assert(false), and add it instead of the NULL.

> +            energies[w*16+g] = log2f(2*(energy*energy));
> +            energy_avg = (energies[w*16+g] + energy_avg)/2;

Not sure I follow the math there. Why the average?

It seems to me that you're trying to compensate for the encoding error
on the first patch (failing to initialize off_pns with -90).

> -                    if (sce->sf_idx[i] < 218 - qstep)
> +                    if (sce->sf_idx[i] < 218 - qstep && sce->band_type[i] < NOISE_BT)
>                          sce->sf_idx[i] += qstep;
>              } else {
>                  for (i = 0; i < 128; i++)
> -                    if (sce->sf_idx[i] > 60 - qstep)
> +                    if (sce->sf_idx[i] > 60 - qstep && sce->band_type[i] < NOISE_BT)
>                          sce->sf_idx[i] -= qstep;

Careful here. You're using band_type[i], but not all of them will have
been initialize to NOISE_BT (ie: consider window groups, only w*16+g
is intialized, with w the first window in the group, the rest are not
initialized.

Probably the easiest solution is to initialize them all to NOISE_BT.

> +            if (freq > 4000.0f && energy <= uplim * 1.52f) {

4000.0f should probably be a constant somewhere.

> +            float freq = (w*16+g)*(avctx->sample_rate/(1024/sce->ics.num_windows)/2);

This is wrong. Should be

start*avctx->sample_rate/(1024/sce->ics.num_windows)/2

More precisely, the window doesn't count for this calculation, you
don't care which window it is, only which coefficient.

And that will be an int, not a float.

Other than that, it looks fine.

You should try to fix the off_pns issue, retest, and resubmit.