AAC Encoder: clipping avoidance

Avoid clipping due to quantization noise to produce audible
artifacts, by detecting near-clipping signals and both attenuating
them a little and encoding escape-encoded bands (usually the
loudest) rounding towards zero instead of nearest, which tends to
decrease overall energy and thus clipping.

Currently fate tests measure numerical error so this change makes
tests using asynth (which are near clipping) report higher error
not less, because of window attenuation. Yet, they sound better,
not worse (albeit subtle, other samples aren't subtle at all).
Only measuring psychoacoustically weighted error would make for
a representative test, so that will be left for a future patch.

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

1 files changed, 30 insertions, 0 deletions

diff --git a/libavcodec/aacpsy.c b/libavcodec/aacpsy.c
index b16f6b9..a5474b9 100644
--- a/libavcodec/aacpsy.c
+++ b/libavcodec/aacpsy.c
@@ -837,6 +837,7 @@ static FFPsyWindowInfo psy_lame_window(FFPsyContext *ctx, const float *audio,
     int grouping     = 0;
     int uselongblock = 1;
     int attacks[AAC_NUM_BLOCKS_SHORT + 1] = { 0 };
+    float clippings[AAC_NUM_BLOCKS_SHORT];
     int i;
     FFPsyWindowInfo wi = { { 0 } };
 
@@ -926,14 +927,35 @@ static FFPsyWindowInfo psy_lame_window(FFPsyContext *ctx, const float *audio,
 
     lame_apply_block_type(pch, &wi, uselongblock);
 
+    /* Calculate input sample maximums and evaluate clipping risk */
+    if (audio) {
+        for (i = 0; i < AAC_NUM_BLOCKS_SHORT; i++) {
+            const float *wbuf = audio + i * AAC_BLOCK_SIZE_SHORT;
+            float max = 0;
+            int j;
+            for (j = 0; j < AAC_BLOCK_SIZE_SHORT; j++)
+                max = FFMAX(max, fabsf(wbuf[j]));
+            clippings[i] = max;
+        }
+    } else {
+        for (i = 0; i < 8; i++)
+            clippings[i] = 0;
+    }
+
     wi.window_type[1] = prev_type;
     if (wi.window_type[0] != EIGHT_SHORT_SEQUENCE) {
+        float clipping = 0.0f;
+
         wi.num_windows  = 1;
         wi.grouping[0]  = 1;
         if (wi.window_type[0] == LONG_START_SEQUENCE)
             wi.window_shape = 0;
         else
             wi.window_shape = 1;
+
+        for (i = 0; i < 8; i++)
+            clipping = FFMAX(clipping, clippings[i]);
+        wi.clipping[0] = clipping;
     } else {
         int lastgrp = 0;
 
@@ -944,6 +966,14 @@ static FFPsyWindowInfo psy_lame_window(FFPsyContext *ctx, const float *audio,
                 lastgrp = i;
             wi.grouping[lastgrp]++;
         }
+
+        for (i = 0; i < 8; i += wi.grouping[i]) {
+            int w;
+            float clipping = 0.0f;
+            for (w = 0; w < wi.grouping[i] && !clipping; w++)
+                clipping = FFMAX(clipping, clippings[i+w]);
+            wi.clipping[i] = clipping;
+        }
     }
 
     /* Determine grouping, based on the location of the first attack, and save for