[FFmpeg-devel] [RFC] AAC Encoder

Kostya kostya.shishkov
Thu Aug 14 07:48:42 CEST 2008


On Wed, Aug 13, 2008 at 04:44:18PM +0200, Michael Niedermayer wrote:
> On Wed, Aug 13, 2008 at 04:42:56PM +0300, Kostya wrote:
> > On Wed, Aug 13, 2008 at 02:57:50PM +0200, Michael Niedermayer wrote:
> [...]
> > 
> > > > 3. based on psy model suggestions, encoder performs windowing and MDCT
> > > 
> > > ok
> > > 
> > > 
> > > > 4. encoder feeds coefficients to psy model
> > > > 5. psy model by some magic determines scalefactors and use them to convert
> > > > coefficients into integer form
> > > > 6. encoder encodes obtained scalefactors and integer coefficients
> > > > 
> > > > There are 11 codebooks for AAC, each designed to code either pairs or quads
> > > > of values with sign coded separately or incorporated into value,
> > > > each has a maximum value limit.
> > > > While it's feasible to find the best encoding (like take raw coeff, quantize
> > > > it and round up or down, then see which vector takes less bits), I feel
> > > > it would be too slow.
> > > 
> > > thats fine, you already have the fast variant implemented i do not suggest
> > > that to be removed, what we need is a high quality variant. The encoder should
> > > be better than other encoders ...
> > > Also as the max value you mentioned is another example of where your code
> > > fails fatally, a single +3 that would sound nearly as good when encoded as +2
> > > could force a less efficient code book to be choosen. Also the +3 could be
> > > encoded as a pulse, i dont remember if your code optimally choose between
> > > pulse and normal codebook encodings?
> > 
> > not optimally, unfortunately, but it can search for pulses and encode them
> > 
> > in any case, here's a new encoder version
> 
> please commit the parts ive ok-ed and/or send a patch without them

done (there were okayed parts only in aacenc.c)
 
> [...]
> -- 
> Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
> 
> Observe your enemies, for they first find out your faults. -- Antisthenes
-------------- next part --------------
--- aacenc.c	2008-08-14 08:44:51.000000000 +0300
+++ /home/kst/Mail/aacenc.c	2008-08-13 16:35:17.000000000 +0300
@@ -245,6 +245,55 @@
     return 0;
 }
 
+static void apply_window_and_mdct(AVCodecContext *avctx, AACEncContext *s, ChannelElement *cpe, short *audio, int channel)
+{
+    int i, j, k;
+    const float * lwindow = cpe->ch[channel].ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_aac_sine_long_1024;
+    const float * swindow = cpe->ch[channel].ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_aac_sine_short_128;
+    const float * pwindow = cpe->ch[channel].ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_aac_sine_short_128;
+
+    if (cpe->ch[channel].ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
+        memcpy(s->output, cpe->ch[channel].saved, sizeof(float)*1024);
+        if(cpe->ch[channel].ics.window_sequence[0] == LONG_STOP_SEQUENCE){
+            memset(s->output, 0, sizeof(s->output[0]) * 448);
+            for(i = 448; i < 576; i++)
+                s->output[i] = cpe->ch[channel].saved[i] * pwindow[i - 448];
+            for(i = 576; i < 704; i++)
+                s->output[i] = cpe->ch[channel].saved[i];
+        }
+        if(cpe->ch[channel].ics.window_sequence[0] != LONG_START_SEQUENCE){
+            j = channel;
+            for (i = 0; i < 1024; i++, j += avctx->channels){
+                s->output[i+1024]         = audio[j] * lwindow[1024 - i - 1];
+                cpe->ch[channel].saved[i] = audio[j] * lwindow[i];
+            }
+        }else{
+            j = channel;
+            for(i = 0; i < 448; i++, j += avctx->channels)
+                s->output[i+1024]         = audio[j];
+            for(i = 448; i < 576; i++, j += avctx->channels)
+                s->output[i+1024]         = audio[j] * swindow[576 - i - 1];
+            memset(s->output+1024+576, 0, sizeof(s->output[0]) * 448);
+            j = channel;
+            for(i = 0; i < 1024; i++, j += avctx->channels)
+                cpe->ch[channel].saved[i] = audio[j];
+        }
+        ff_mdct_calc(&s->mdct1024, cpe->ch[channel].coeffs, s->output, s->tmp);
+    }else{
+        j = channel;
+        for (k = 0; k < 1024; k += 128) {
+            for(i = 448 + k; i < 448 + k + 256; i++)
+                s->output[i - 448 - k] = (i < 1024) ? cpe->ch[channel].saved[i] : audio[channel + (i-1024)*avctx->channels] / 512.0;
+            s->dsp.vector_fmul        (s->output,     k ?  swindow : pwindow, 128);
+            s->dsp.vector_fmul_reverse(s->output+128, s->output+128, swindow, 128);
+            ff_mdct_calc(&s->mdct128, cpe->ch[channel].coeffs + k, s->output, s->tmp);
+        }
+        j = channel;
+        for(i = 0; i < 1024; i++, j += avctx->channels)
+            cpe->ch[channel].saved[i] = audio[j];
+    }
+}
+
 /**
  * Encode ics_info element.
  * @see Table 4.6 (syntax of ics_info)
@@ -268,6 +317,327 @@
 }
 
 /**
+ * Encode MS data.
+ * @see 4.6.8.1 "Joint Coding - M/S Stereo"
+ */
+static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
+{
+    int i, w;
+
+    put_bits(pb, 2, cpe->ms.present);
+    if(cpe->ms.present == 1)
+        for(w = 0; w < cpe->ch[0].ics.num_windows; w++){
+            if(cpe->ch[0].ics.group_len[w]) continue;
+            for(i = 0; i < cpe->ch[0].ics.max_sfb; i++)
+                put_bits(pb, 1, cpe->ms.mask[w][i]);
+        }
+}
+
+/**
+ * Scan scalefactor band and determine optimal codebook for it.
+ *
+ * @param s       encoder context
+ * @param cpe     channel element
+ * @param channel channel number inside channel pair
+ * @param win     window group start number
+ * @param band    scalefactor band to analyze
+ * @param start   scalefactor band position in spectral coefficients
+ * @param size    scalefactor band size
+ */
+static int determine_section_info(AACEncContext *s, ChannelElement *cpe, int channel, int win, int band, int start, int size)
+{
+    int i, j, w;
+    int maxval, sign;
+    int score, best, cb, bestcb, dim, idx, start2;
+
+    maxval = 0;
+    sign = 0;
+    w = win;
+    start2 = start;
+    do{
+        for(i = start2; i < start2 + size; i++){
+            maxval = FFMAX(maxval, FFABS(cpe->ch[channel].icoefs[i]));
+            if(cpe->ch[channel].icoefs[i] < 0) sign = 1;
+        }
+        w++;
+        start2 += 128;
+    }while(w < cpe->ch[channel].ics.num_windows && cpe->ch[channel].ics.group_len[w]);
+
+    if(maxval > 12) return 11;
+    if(!maxval) return 0;
+
+    for(cb = 0; cb < 12; cb++)
+        if(aac_cb_info[cb].maxval >= maxval)
+            break;
+    best = INT_MAX;
+    bestcb = 11;
+    for(; cb < 12; cb++){
+        score = 0;
+        dim = (aac_cb_info[cb].flags & CB_PAIRS) ? 2 : 4;
+        if(!band || cpe->ch[channel].band_type[win][band - 1] != cb)
+            score += 9; //that's for new codebook entry
+        w = win;
+        start2 = start;
+        if(aac_cb_info[cb].flags & CB_UNSIGNED){
+            do{
+                for(i = start2; i < start2 + size; i += dim){
+                    idx = 0;
+                    for(j = 0; j < dim; j++)
+                        idx = idx * aac_cb_info[cb].maxval + FFABS(cpe->ch[channel].icoefs[i+j]);
+                    score += ff_aac_spectral_bits[aac_cb_info[cb].cb_num][idx];
+                    for(j = 0; j < dim; j++)
+                        if(cpe->ch[channel].icoefs[i+j])
+                            score++;
+                }
+                w++;
+                start2 += 128;
+            }while(w < cpe->ch[channel].ics.num_windows && cpe->ch[channel].ics.group_len[w]);
+        }else{
+            do{
+                for(i = start2; i < start2 + size; i += dim){
+                    idx = 0;
+                    for(j = 0; j < dim; j++)
+                        idx = idx * (aac_cb_info[cb].maxval*2 + 1) + cpe->ch[channel].icoefs[i+j] + aac_cb_info[cb].maxval;
+                    score += ff_aac_spectral_bits[aac_cb_info[cb].cb_num][idx];
+                }
+                w++;
+                start2 += 128;
+            }while(w < cpe->ch[channel].ics.num_windows && cpe->ch[channel].ics.group_len[w]);
+        }
+        if(score < best){
+            best = score;
+            bestcb = cb;
+        }
+    }
+    return bestcb;
+}
+
+/**
+ * Encode one scalefactor band with selected codebook.
+ */
+static void encode_band_coeffs(AACEncContext *s, ChannelElement *cpe, int channel, int start, int size, int cb)
+{
+    const uint8_t  *bits  = ff_aac_spectral_bits [aac_cb_info[cb].cb_num];
+    const uint16_t *codes = ff_aac_spectral_codes[aac_cb_info[cb].cb_num];
+    const int dim = (aac_cb_info[cb].flags & CB_PAIRS) ? 2 : 4;
+    int i, j, idx;
+
+    if(!bits) return;
+
+    if(aac_cb_info[cb].flags & CB_ESCAPE){
+        for(i = start; i < start + size; i += dim){
+            idx = 0;
+            for(j = 0; j < dim; j++)
+                idx = idx*17 + FFMIN(FFABS(cpe->ch[channel].icoefs[i+j]), 16);
+            put_bits(&s->pb, bits[idx], codes[idx]);
+            //output signs
+            for(j = 0; j < dim; j++)
+                if(cpe->ch[channel].icoefs[i+j])
+                    put_bits(&s->pb, 1, cpe->ch[channel].icoefs[i+j] < 0);
+            //output escape values
+            for(j = 0; j < dim; j++)
+                if(FFABS(cpe->ch[channel].icoefs[i+j]) > 15){
+                    int l = av_log2(FFABS(cpe->ch[channel].icoefs[i+j]));
+
+                    put_bits(&s->pb, l - 4 + 1, (1 << (l - 4 + 1)) - 2);
+                    put_bits(&s->pb, l, FFABS(cpe->ch[channel].icoefs[i+j]) & ((1 << l) - 1));
+                }
+        }
+    }else if(aac_cb_info[cb].flags & CB_UNSIGNED){
+        for(i = start; i < start + size; i += dim){
+            idx = 0;
+            for(j = 0; j < dim; j++)
+                idx = idx * (aac_cb_info[cb].maxval + 1) + FFABS(cpe->ch[channel].icoefs[i+j]);
+            put_bits(&s->pb, bits[idx], codes[idx]);
+            //output signs
+            for(j = 0; j < dim; j++)
+                if(cpe->ch[channel].icoefs[i+j])
+                    put_bits(&s->pb, 1, cpe->ch[channel].icoefs[i+j] < 0);
+        }
+    }else{
+        for(i = start; i < start + size; i += dim){
+            idx = 0;
+            for(j = 0; j < dim; j++)
+                idx = idx * (aac_cb_info[cb].maxval*2 + 1) + cpe->ch[channel].icoefs[i+j] + aac_cb_info[cb].maxval;
+            put_bits(&s->pb, bits[idx], codes[idx]);
+        }
+    }
+}
+
+/**
+ * Encode scalefactor band coding type.
+ */
+static void encode_band_info(AVCodecContext *avctx, AACEncContext *s, ChannelElement *cpe, int channel)
+{
+    int i, w;
+    int bits = cpe->ch[channel].ics.num_windows == 1 ? 5 : 3;
+    int esc = (1 << bits) - 1;
+    int count;
+
+    for(w = 0; w < cpe->ch[channel].ics.num_windows; w++){
+        if(cpe->ch[channel].ics.group_len[w]) continue;
+        count = 0;
+        for(i = 0; i < cpe->ch[channel].ics.max_sfb; i++){
+            if(!i || cpe->ch[channel].band_type[w][i] != cpe->ch[channel].band_type[w][i-1]){
+                if(count){
+                    while(count >= esc){
+                        put_bits(&s->pb, bits, esc);
+                        count -= esc;
+                    }
+                    put_bits(&s->pb, bits, count);
+                }
+                put_bits(&s->pb, 4, cpe->ch[channel].band_type[w][i]);
+                count = 1;
+            }else
+                count++;
+        }
+        if(count){
+            while(count >= esc){
+                put_bits(&s->pb, bits, esc);
+                count -= esc;
+            }
+            put_bits(&s->pb, bits, count);
+        }
+    }
+}
+
+/**
+ * Encode scalefactors.
+ */
+static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s, ChannelElement *cpe, int channel)
+{
+    int off = cpe->ch[channel].mixing_gain, diff;
+    int i, w;
+
+    for(w = 0; w < cpe->ch[channel].ics.num_windows; w++){
+        if(cpe->ch[channel].ics.group_len[w]) continue;
+        for(i = 0; i < cpe->ch[channel].ics.max_sfb; i++){
+            if(!cpe->ch[channel].zeroes[w][i]){
+                diff = cpe->ch[channel].sf_idx[w][i] - off + SCALE_DIFF_ZERO;
+                if(diff < 0 || diff > 120) av_log(avctx, AV_LOG_ERROR, "Scalefactor difference is too big to be coded\n");
+                off = cpe->ch[channel].sf_idx[w][i];
+                put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
+            }
+        }
+    }
+}
+
+/**
+ * Encode pulse data.
+ */
+static void encode_pulses(AVCodecContext *avctx, AACEncContext *s, ChannelElement *cpe, int channel)
+{
+    int i;
+
+    put_bits(&s->pb, 1, !!cpe->ch[channel].pulse.num_pulse);
+    if(!cpe->ch[channel].pulse.num_pulse) return;
+
+    put_bits(&s->pb, 2, cpe->ch[channel].pulse.num_pulse - 1);
+    put_bits(&s->pb, 6, cpe->ch[channel].pulse.start);
+    for(i = 0; i < cpe->ch[channel].pulse.num_pulse; i++){
+        put_bits(&s->pb, 5, cpe->ch[channel].pulse.offset[i]);
+        put_bits(&s->pb, 4, cpe->ch[channel].pulse.amp[i]);
+    }
+}
+
+/**
+ * Encode temporal noise shaping data.
+ */
+static void encode_tns_data(AVCodecContext *avctx, AACEncContext *s, ChannelElement *cpe, int channel)
+{
+    int i, w;
+
+    put_bits(&s->pb, 1, cpe->ch[channel].tns.present);
+    if(!cpe->ch[channel].tns.present) return;
+    if(cpe->ch[channel].ics.window_sequence[0] == EIGHT_SHORT_SEQUENCE){
+        for(w = 0; w < cpe->ch[channel].ics.num_windows; w++){
+            put_bits(&s->pb, 1, cpe->ch[channel].tns.n_filt[w]);
+            if(!cpe->ch[channel].tns.n_filt[w]) continue;
+            put_bits(&s->pb, 1, cpe->ch[channel].tns.coef_res[w] - 3);
+            put_bits(&s->pb, 4, cpe->ch[channel].tns.length[w][0]);
+            put_bits(&s->pb, 3, cpe->ch[channel].tns.order[w][0]);
+            if(cpe->ch[channel].tns.order[w][0]){
+                put_bits(&s->pb, 1, cpe->ch[channel].tns.direction[w][0]);
+                put_bits(&s->pb, 1, cpe->ch[channel].tns.coef_compress[w][0]);
+                for(i = 0; i < cpe->ch[channel].tns.order[w][0]; i++)
+                     put_bits(&s->pb, cpe->ch[channel].tns.coef_len[w][0], cpe->ch[channel].tns.coef[w][0][i]);
+            }
+        }
+    }else{
+        put_bits(&s->pb, 1, cpe->ch[channel].tns.n_filt[0]);
+        if(!cpe->ch[channel].tns.n_filt[0]) return;
+        put_bits(&s->pb, 1, cpe->ch[channel].tns.coef_res[0] - 3);
+        for(w = 0; w < cpe->ch[channel].tns.n_filt[0]; w++){
+            put_bits(&s->pb, 6, cpe->ch[channel].tns.length[0][w]);
+            put_bits(&s->pb, 5, cpe->ch[channel].tns.order[0][w]);
+            if(cpe->ch[channel].tns.order[0][w]){
+                put_bits(&s->pb, 1, cpe->ch[channel].tns.direction[0][w]);
+                put_bits(&s->pb, 1, cpe->ch[channel].tns.coef_compress[0][w]);
+                for(i = 0; i < cpe->ch[channel].tns.order[0][w]; i++)
+                     put_bits(&s->pb, cpe->ch[channel].tns.coef_len[0][w], cpe->ch[channel].tns.coef[0][w][i]);
+            }
+        }
+    }
+}
+
+/**
+ * Encode spectral coefficients processed by psychoacoustic model.
+ */
+static void encode_spectral_coeffs(AVCodecContext *avctx, AACEncContext *s, ChannelElement *cpe, int channel)
+{
+    int start, i, w, w2;
+
+    for(w = 0; w < cpe->ch[channel].ics.num_windows; w++){
+        if(cpe->ch[channel].ics.group_len[w]) continue;
+        start = 0;
+        for(i = 0; i < cpe->ch[channel].ics.max_sfb; i++){
+            if(cpe->ch[channel].zeroes[w][i]){
+                start += cpe->ch[channel].ics.swb_sizes[i];
+                continue;
+            }
+            w2 = w;
+            do{
+                encode_band_coeffs(s, cpe, channel, start + w2*128, cpe->ch[channel].ics.swb_sizes[i], cpe->ch[channel].band_type[w][i]);
+                w2++;
+            }while(w2 < cpe->ch[channel].ics.num_windows && cpe->ch[channel].ics.group_len[w2]);
+            start += cpe->ch[channel].ics.swb_sizes[i];
+        }
+    }
+}
+
+/**
+ * Encode one channel of audio data.
+ */
+static int encode_individual_channel(AVCodecContext *avctx, ChannelElement *cpe, int channel)
+{
+    AACEncContext *s = avctx->priv_data;
+    int i, g, w;
+
+    for(w = 0; w < cpe->ch[channel].ics.num_windows; w++){
+        i = w << 7;
+        if(cpe->ch[channel].ics.group_len[w]) continue;
+        for(g = 0; g < cpe->ch[channel].ics.max_sfb; g++){
+            if(!cpe->ch[channel].zeroes[w][g]){
+                cpe->ch[channel].band_type[w][g] = determine_section_info(s, cpe, channel, w, g, i, cpe->ch[channel].ics.swb_sizes[g]);
+                cpe->ch[channel].zeroes[w][g] = !cpe->ch[channel].band_type[w][g];
+            }else
+                cpe->ch[channel].band_type[w][g] = 0;
+            i += cpe->ch[channel].ics.swb_sizes[g];
+        }
+    }
+
+    put_bits(&s->pb, 8, cpe->ch[channel].mixing_gain); //global gain
+    if(!cpe->common_window) put_ics_info(avctx, &cpe->ch[channel].ics);
+    encode_band_info(avctx, s, cpe, channel);
+    encode_scale_factors(avctx, s, cpe, channel);
+    encode_pulses(avctx, s, cpe, channel);
+    encode_tns_data(avctx, s, cpe, channel);
+    put_bits(&s->pb, 1, 0); //ssr
+    encode_spectral_coeffs(avctx, s, cpe, channel);
+    return 0;
+}
+
+/**
  * Write some auxiliary information about the created AAC file.
  */
 static void put_bitstream_info(AVCodecContext *avctx, AACEncContext *s, const char *name)
@@ -287,6 +657,80 @@
     put_bits(&s->pb, 12 - padbits, 0);
 }
 
+static int aac_encode_frame(AVCodecContext *avctx,
+                            uint8_t *frame, int buf_size, void *data)
+{
+    AACEncContext *s = avctx->priv_data;
+    int16_t *samples = s->samples, *samples2, *la;
+    ChannelElement *cpe;
+    int i, j, chans, tag, start_ch;
+    const uint8_t *chan_map = aac_chan_configs[avctx->channels-1];
+    int chan_el_counter[4];
+
+    if(s->last_frame)
+        return 0;
+    if(data){
+        if((s->psy.flags & PSY_MODEL_NO_PREPROC) == PSY_MODEL_NO_PREPROC){
+            memcpy(s->samples + 1024 * avctx->channels, data, 1024 * avctx->channels * sizeof(s->samples[0]));
+        }else{
+            start_ch = 0;
+            samples2 = s->samples + 1024 * avctx->channels;
+            for(i = 0; i < chan_map[0]; i++){
+                tag = chan_map[i+1];
+                chans = tag == ID_CPE ? 2 : 1;
+                ff_aac_psy_preprocess(&s->psy, (uint16_t*)data + start_ch, samples2 + start_ch, i, tag);
+                start_ch += chans;
+            }
+        }
+    }
+    if(!avctx->frame_number){
+        memmove(s->samples, s->samples + 1024 * avctx->channels, 1024 * avctx->channels * sizeof(s->samples[0]));
+        return 0;
+    }
+
+    init_put_bits(&s->pb, frame, buf_size*8);
+    if(avctx->frame_number==1 && !(avctx->flags & CODEC_FLAG_BITEXACT)){
+        put_bitstream_info(avctx, s, LIBAVCODEC_IDENT);
+    }
+    start_ch = 0;
+    memset(chan_el_counter, 0, sizeof(chan_el_counter));
+    for(i = 0; i < chan_map[0]; i++){
+        tag = chan_map[i+1];
+        chans = tag == ID_CPE ? 2 : 1;
+        cpe = &s->cpe[i];
+        samples2 = samples + start_ch;
+        la = samples2 + 1024 * avctx->channels + start_ch;
+        if(!data) la = NULL;
+        ff_aac_psy_suggest_window(&s->psy, samples2, la, i, tag, cpe);
+        for(j = 0; j < chans; j++){
+            apply_window_and_mdct(avctx, s, cpe, samples2, j);
+        }
+        ff_aac_psy_analyze(&s->psy, i, tag, cpe);
+        put_bits(&s->pb, 3, tag);
+        put_bits(&s->pb, 4, chan_el_counter[tag]++);
+        if(chans == 2){
+            put_bits(&s->pb, 1, cpe->common_window);
+            if(cpe->common_window){
+                put_ics_info(avctx, &cpe->ch[0].ics);
+                encode_ms_info(&s->pb, cpe);
+            }
+        }
+        for(j = 0; j < chans; j++){
+            encode_individual_channel(avctx, cpe, j);
+        }
+        start_ch += chans;
+    }
+
+    put_bits(&s->pb, 3, ID_END);
+    flush_put_bits(&s->pb);
+    avctx->frame_bits = put_bits_count(&s->pb);
+
+    if(!data)
+        s->last_frame = 1;
+    memmove(s->samples, s->samples + 1024 * avctx->channels, 1024 * avctx->channels * sizeof(s->samples[0]));
+    return put_bits_count(&s->pb)>>3;
+}
+
 static av_cold int aac_encode_end(AVCodecContext *avctx)
 {
     AACEncContext *s = avctx->priv_data;
-------------- next part --------------
/*
 * AAC encoder psychoacoustic model
 * Copyright (C) 2008 Konstantin Shishkov
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#ifndef FFMPEG_AACPSY_H
#define FFMPEG_AACPSY_H

#include "avcodec.h"
#include "aac.h"
#include "lowpass.h"

enum AACPsyModelType{
    AAC_PSY_NULL,              ///< do nothing with frequencies
    AAC_PSY_NULL8,             ///< do nothing with frequencies but work with short windows
    AAC_PSY_3GPP,              ///< model following recommendations from 3GPP TS 26.403

    AAC_NB_PSY_MODELS          ///< total number of psychoacoustic models, since it's not a part of the ABI new models can be added freely
};

enum AACPsyModelMode{
    PSY_MODE_CBR,              ///< follow bitrate as closely as possible
    PSY_MODE_ABR,              ///< try to achieve bitrate but actual bitrate may differ significantly
    PSY_MODE_QUALITY,          ///< try to achieve set quality instead of bitrate
};

#define PSY_MODEL_MODE_MASK  0x0000000F ///< bit fields for storing mode (CBR, ABR, VBR)
#define PSY_MODEL_NO_PULSE   0x00000010 ///< disable pulse searching
#define PSY_MODEL_NO_SWITCH  0x00000020 ///< disable window switching
#define PSY_MODEL_NO_ST_ATT  0x00000040 ///< disable stereo attenuation
#define PSY_MODEL_NO_LOWPASS 0x00000080 ///< disable low-pass filtering

#define PSY_MODEL_NO_PREPROC (PSY_MODEL_NO_ST_ATT | PSY_MODEL_NO_LOWPASS)

#define PSY_MODEL_MODE(a)  ((a) & PSY_MODEL_MODE_MASK)

/**
 * context used by psychoacoustic model
 */
typedef struct AACPsyContext {
    AVCodecContext *avctx;            ///< encoder context

    int flags;                        ///< model flags
    const uint8_t *bands1024;         ///< scalefactor band sizes for long (1024 samples) frame
    int num_bands1024;                ///< number of scalefactor bands for long frame
    const uint8_t *bands128;          ///< scalefactor band sizes for short (128 samples) frame
    int num_bands128;                 ///< number of scalefactor bands for short frame

    const struct AACPsyModel *model;  ///< pointer to the psychoacoustic model implementation
    void* model_priv_data;            ///< psychoacoustic model implementation private data

    float stereo_att;                 ///< stereo attenuation factor
    LPFilterCoeffs lp_coeffs;         ///< lowpass filter coefficients
    LPFilterState *lp_state;          ///< lowpass filter state
}AACPsyContext;

typedef struct AACPsyModel {
    const char *name;
    int   (*init)   (AACPsyContext *apc, int elements);
    void  (*window) (AACPsyContext *apc, int16_t *audio, int16_t *la, int tag, int type, ChannelElement *cpe);
    void  (*process)(AACPsyContext *apc, int tag, int type, ChannelElement *cpe);
    void  (*end)    (AACPsyContext *apc);
}AACPsyModel;

/**
 * Initialize psychoacoustic model.
 *
 * @param ctx           model context
 * @param avctx         codec context
 * @param model         model implementation that will be used
 * @param elements      number of channel elements (single channel or channel pair) to handle by  model
 * @param flags         model flags, may be ignored by model if unsupported
 * @param bands1024     scalefactor band lengths for long (1024 samples) frame
 * @param num_bands1024 number of scalefactor bands for long frame
 * @param bands128      scalefactor band lengths for short (128 samples) frame
 * @param num_bands128  number of scalefactor bands for short frame
 *
 * @return zero if successful, a negative value if not
 */
int ff_aac_psy_init(AACPsyContext *ctx, AVCodecContext *avctx,
                    enum AACPsyModelType model, int elements, int flags,
                    const uint8_t *bands1024, int num_bands1024,
                    const uint8_t *bands128,  int num_bands128);

/**
 * Preprocess audio frame in order to compress it better.
 *
 * @param ctx   model context
 * @param audio samples to preprocess
 * @param dest  place to put filtered samples
 * @param tag   number of channel element to analyze
 * @param type  channel element type (e.g. ID_SCE or ID_CPE)
 */
void ff_aac_psy_preprocess(AACPsyContext *ctx, int16_t *audio, int16_t *dest, int tag, int type);

/**
 * Set window sequence and related parameters for channel element.
 *
 * @param ctx   model context
 * @param audio samples for the current frame
 * @param la    lookahead samples (NULL when unavailable)
 * @param tag   number of channel element to analyze
 * @param type  channel element type (e.g. ID_SCE or ID_CPE)
 * @param cpe   pointer to the current channel element
 */
void ff_aac_psy_suggest_window(AACPsyContext *ctx, int16_t *audio, int16_t *la, int tag, int type, ChannelElement *cpe);

/**
 * Perform psychoacoustic analysis and output coefficients in integer form
 * along with scalefactors, M/S flags, etc.
 *
 * @param ctx   model context
 * @param tag   number of channel element to analyze
 * @param type  channel element type (e.g. ID_SCE or ID_CPE)
 * @param cpe   pointer to the current channel element
 */
void ff_aac_psy_analyze(AACPsyContext *ctx, int tag, int type, ChannelElement *cpe);

/**
 * Cleanup model context at the end.
 *
 * @param ctx model context
 */
void ff_aac_psy_end(AACPsyContext *ctx);

#endif /* FFMPEG_AACPSY_H */

-------------- next part --------------
/*
 * AAC encoder psychoacoustic model
 * Copyright (C) 2008 Konstantin Shishkov
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file aacpsy.c
 * AAC encoder psychoacoustic model
 */

#include "avcodec.h"
#include "aacpsy.h"
#include "aactab.h"

/***********************************
 *              TODOs:
 * General:
 * better audio preprocessing (add DC highpass filter?)
 * more psy models
 **********************************/

/**
 * Convert coefficients to integers.
 * @return sum of coefficients
 * @see 3GPP TS26.403 5.6.2 "Scalefactor determination"
 */
static inline int convert_coeffs(float *in, int *out, int size, int scale_idx)
{
    int i, sign, sum = 0;
    for(i = 0; i < size; i++){
        sign = in[i] > 0.0;
        out[i] = (int)(pow(FFABS(in[i]) * ff_aac_pow2sf_tab[200 - scale_idx + SCALE_ONE_POS - SCALE_DIV_512], 0.75) + 0.4054);
        out[i] = av_clip(out[i], 0, 8191);
        sum += out[i];
        if(sign) out[i] = -out[i];
    }
    return sum;
}

static inline float unquant(int q, int scale_idx){
    return (FFABS(q) * cbrt(q*1.0)) * ff_aac_pow2sf_tab[200 + scale_idx - SCALE_ONE_POS];
}
static inline float calc_distortion(float *c, int size, int scale_idx)
{
    int i;
    int q;
    float coef, unquant, sum = 0.0f;
    for(i = 0; i < size; i++){
        coef = FFABS(c[i]);
        q = (int)(pow(FFABS(coef) * ff_aac_pow2sf_tab[200 - scale_idx + SCALE_ONE_POS - SCALE_DIV_512], 0.75) + 0.4054);
        q = av_clip(q, 0, 8191);
        unquant = (q * cbrt(q)) * ff_aac_pow2sf_tab[200 + scale_idx - SCALE_ONE_POS + SCALE_DIV_512];
        sum += (coef - unquant) * (coef - unquant);
    }
    return sum;
}

/**
 * Produce integer coefficients from scalefactors provided by the model.
 */
static void psy_create_output(AACPsyContext *apc, ChannelElement *cpe, int chans, int search_pulses)
{
    int i, w, w2, g, ch;
    int start, sum, maxsfb, cmaxsfb;
    int pulses, poff[4], pamp[4];

    for(ch = 0; ch < chans; ch++){
        start = 0;
        maxsfb = 0;
        cpe->ch[ch].pulse.num_pulse = 0;
        for(w = 0; w < cpe->ch[ch].ics.num_windows; w++){
            for(g = 0; g < cpe->ch[ch].ics.num_swb; g++){
                sum = 0;
                //apply M/S
                if(!ch && cpe->ms.mask[w][g]){
                    for(i = 0; i < cpe->ch[ch].ics.swb_sizes[g]; i++){
                        cpe->ch[0].coeffs[start+i] = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) / 2.0;
                        cpe->ch[1].coeffs[start+i] =  cpe->ch[0].coeffs[start+i] - cpe->ch[1].coeffs[start+i];
                    }
                }
                if(!cpe->ch[ch].zeroes[w][g])
                    sum = convert_coeffs(cpe->ch[ch].coeffs + start, cpe->ch[ch].icoefs + start, cpe->ch[ch].ics.swb_sizes[g], cpe->ch[ch].sf_idx[w][g]);
                else
                    memset(cpe->ch[ch].icoefs + start, 0, cpe->ch[ch].ics.swb_sizes[g] * sizeof(cpe->ch[0].icoefs[0]));
                cpe->ch[ch].zeroes[w][g] = !sum;
                //try finding pulses
                if(search_pulses && cpe->ch[ch].ics.num_windows == 1 && !cpe->ch[ch].pulse.num_pulse){
                    pulses = 0;
                    memset(poff,0,sizeof(poff));
                    memset(pamp,0,sizeof(pamp));
                    for(i = 0; i < cpe->ch[ch].ics.swb_sizes[g]; i++){
                        if(pulses > 4 || (pulses && i > cpe->ch[ch].pulse.offset[pulses-1] - 31)) break;
                        if(FFABS(cpe->ch[ch].icoefs[start+i]) > 4 && pulses < 4){
                            poff[pulses] = i;
                            pamp[pulses] = FFMIN(FFABS(cpe->ch[ch].icoefs[start+i]) - 1, 15);
                            pulses++;
                        }
                    }
                    if(pulses){
                        cpe->ch[ch].pulse.start = g;
                        cpe->ch[ch].pulse.num_pulse = pulses;
                        for(i = 0; i < pulses; i++){
                            cpe->ch[ch].pulse.amp[i] = pamp[i];
                            cpe->ch[ch].pulse.offset[i] = i ? poff[i] - poff[i-1] : poff[0];

                            if(cpe->ch[ch].icoefs[start+poff[i]] > 0)
                                cpe->ch[ch].icoefs[start+poff[i]] -= pamp[i];
                            else
                                cpe->ch[ch].icoefs[start+poff[i]] += pamp[i];
                        }
                    }
                }
                start += cpe->ch[ch].ics.swb_sizes[g];
            }
            for(cmaxsfb = cpe->ch[ch].ics.num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w][cmaxsfb-1]; cmaxsfb--);
            maxsfb = FFMAX(maxsfb, cmaxsfb);
        }
        cpe->ch[ch].ics.max_sfb = maxsfb;

        //adjust zero bands for window groups
        for(w = 0; w < cpe->ch[ch].ics.num_windows; w++){
            if(cpe->ch[ch].ics.group_len[w]) continue;
            for(g = 0; g < cpe->ch[ch].ics.max_sfb; g++){
                i = 1;
                w2 = w;
                do{
                    if(!cpe->ch[ch].zeroes[w2][g]){
                        i = 0;
                        break;
                    }
                    w2++;
                }while(w2 < cpe->ch[ch].ics.num_windows && cpe->ch[ch].ics.group_len[w2]);
                cpe->ch[ch].zeroes[w][g] = i;
            }
        }
    }

    if(chans > 1 && cpe->common_window){
        int msc = 0;
        cpe->ch[0].ics.max_sfb = FFMAX(cpe->ch[0].ics.max_sfb, cpe->ch[1].ics.max_sfb);
        cpe->ch[1].ics.max_sfb = cpe->ch[0].ics.max_sfb;
        for(w = 0; w < cpe->ch[0].ics.num_windows; w++)
            for(i = 0; i < cpe->ch[0].ics.max_sfb; i++)
                if(cpe->ms.mask[w][i]) msc++;
        if(msc == 0 || cpe->ch[0].ics.max_sfb == 0) cpe->ms.present = 0;
        else cpe->ms.present = msc < cpe->ch[0].ics.max_sfb ? 1 : 2;
    }
}

static void psy_null_window(AACPsyContext *apc, int16_t *audio, int16_t *la, int tag, int type, ChannelElement *cpe)
{
    int ch;
    int chans = type == ID_CPE ? 2 : 1;

    for(ch = 0; ch < chans; ch++){
        cpe->ch[ch].ics.window_sequence[0] = ONLY_LONG_SEQUENCE;
        cpe->ch[ch].ics.use_kb_window[0] = 1;
        cpe->ch[ch].ics.num_windows = 1;
        cpe->ch[ch].ics.swb_sizes = apc->bands1024;
        cpe->ch[ch].ics.num_swb = apc->num_bands1024;
        cpe->ch[ch].ics.group_len[0] = 0;
    }
    cpe->common_window = cpe->ch[0].ics.use_kb_window[0] == cpe->ch[1].ics.use_kb_window[0];
}

static void psy_null_process(AACPsyContext *apc, int tag, int type, ChannelElement *cpe)
{
    int start;
    int ch, g, i;
    int minscale;
    int chans = type == ID_CPE ? 2 : 1;

    for(ch = 0; ch < chans; ch++){
        start = 0;
        for(g = 0; g < apc->num_bands1024; g++){
            float energy = 0.0f, ffac = 0.0f, thr, dist;

            for(i = 0; i < apc->bands1024[g]; i++){
                energy += cpe->ch[ch].coeffs[start+i]*cpe->ch[ch].coeffs[start+i];
                ffac += sqrt(FFABS(cpe->ch[ch].coeffs[start+i]));
            }
            thr = energy * 0.001258925f;
            cpe->ch[ch].sf_idx[ch][g] = 136;
            cpe->ch[ch].zeroes[ch][g] = (energy == 0.0);
            if(cpe->ch[ch].zeroes[ch][g]) continue;
            minscale = (int)(2.66667 * (log2(6.75*thr) - log2(ffac)));
            cpe->ch[ch].sf_idx[ch][g] = SCALE_ONE_POS - minscale;
            while(cpe->ch[ch].sf_idx[ch][g] > 3){
                dist = calc_distortion(cpe->ch[ch].coeffs + start, apc->bands1024[g], cpe->ch[ch].sf_idx[ch][g]);
                if(dist < thr) break;
                cpe->ch[ch].sf_idx[ch][g] -= 3;
            }
        }
    }
    for(ch = 0; ch < chans; ch++){
        minscale = 255;
        for(g = 0; g < apc->num_bands1024; g++)
            if(!cpe->ch[ch].zeroes[0][g])
                minscale = FFMIN(minscale, cpe->ch[ch].sf_idx[0][g]);
        cpe->ch[ch].mixing_gain = minscale;
        for(g = 0; g < apc->num_bands1024; g++)
            if(!cpe->ch[ch].zeroes[0][g])
                cpe->ch[ch].sf_idx[0][g] = FFMIN(minscale + SCALE_MAX_DIFF, cpe->ch[ch].sf_idx[0][g]);
    }
    psy_create_output(apc, cpe, chans, 1);
}

static void psy_null8_window(AACPsyContext *apc, int16_t *audio, int16_t *la, int tag, int type, ChannelElement *cpe)
{
    int ch, i;
    int chans = type == ID_CPE ? 2 : 1;

    for(ch = 0; ch < chans; ch++){
        int prev_seq = cpe->ch[ch].ics.window_sequence[1];
        cpe->ch[ch].ics.use_kb_window[1] = cpe->ch[ch].ics.use_kb_window[0];
        cpe->ch[ch].ics.window_sequence[1] = cpe->ch[ch].ics.window_sequence[0];
        switch(cpe->ch[ch].ics.window_sequence[0]){
        case ONLY_LONG_SEQUENCE:   if(prev_seq == ONLY_LONG_SEQUENCE)cpe->ch[ch].ics.window_sequence[0] = LONG_START_SEQUENCE;   break;
        case LONG_START_SEQUENCE:  cpe->ch[ch].ics.window_sequence[0] = EIGHT_SHORT_SEQUENCE; break;
        case EIGHT_SHORT_SEQUENCE: if(prev_seq == EIGHT_SHORT_SEQUENCE)cpe->ch[ch].ics.window_sequence[0] = LONG_STOP_SEQUENCE;  break;
        case LONG_STOP_SEQUENCE:   cpe->ch[ch].ics.window_sequence[0] = ONLY_LONG_SEQUENCE;   break;
        }

        if(cpe->ch[ch].ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE){
            cpe->ch[ch].ics.use_kb_window[0] = 1;
            cpe->ch[ch].ics.num_windows = 1;
            cpe->ch[ch].ics.swb_sizes = apc->bands1024;
            cpe->ch[ch].ics.num_swb = apc->num_bands1024;
            cpe->ch[ch].ics.group_len[0] = 0;
        }else{
            cpe->ch[ch].ics.use_kb_window[0] = 1;
            cpe->ch[ch].ics.num_windows = 8;
            cpe->ch[ch].ics.swb_sizes = apc->bands128;
            cpe->ch[ch].ics.num_swb = apc->num_bands128;
            for(i = 0; i < cpe->ch[ch].ics.num_windows; i++)
                cpe->ch[ch].ics.group_len[i] = i & 1;
        }
    }
    cpe->common_window = cpe->ch[0].ics.use_kb_window[0] == cpe->ch[1].ics.use_kb_window[0];
}

static void psy_null8_process(AACPsyContext *apc, int tag, int type, ChannelElement *cpe)
{
    int start;
    int w, ch, g, i;
    int chans = type == ID_CPE ? 2 : 1;

    //detect M/S
    if(chans > 1 && cpe->common_window){
        start = 0;
        for(w = 0; w < cpe->ch[0].ics.num_windows; w++){
            for(g = 0; g < cpe->ch[0].ics.num_swb; g++){
                float diff = 0.0f;

                for(i = 0; i < cpe->ch[0].ics.swb_sizes[g]; i++)
                    diff += fabs(cpe->ch[0].coeffs[start+i] - cpe->ch[1].coeffs[start+i]);
                cpe->ms.mask[w][g] = diff == 0.0;
            }
        }
    }
    for(ch = 0; ch < chans; ch++){
        cpe->ch[ch].mixing_gain = SCALE_ONE_POS;
        for(w = 0; w < cpe->ch[ch].ics.num_windows; w++){
            for(g = 0; g < cpe->ch[ch].ics.num_swb; g++){
                cpe->ch[ch].sf_idx[w][g] = SCALE_ONE_POS;
                cpe->ch[ch].zeroes[w][g] = 0;
            }
        }
    }
    psy_create_output(apc, cpe, chans, 0);
}

static const AACPsyModel psy_models[AAC_NB_PSY_MODELS] =
{
    {
       "Null model",
        NULL,
        psy_null_window,
        psy_null_process,
        NULL,
    },
    {
       "Null model - short windows",
        NULL,
        psy_null8_window,
        psy_null8_process,
        NULL,
    },
};

int av_cold ff_aac_psy_init(AACPsyContext *ctx, AVCodecContext *avctx,
                            enum AACPsyModelType model, int elements, int flags,
                            const uint8_t *bands1024, int num_bands1024,
                            const uint8_t *bands128,  int num_bands128)
{
    int i;

    if(model >= AAC_NB_PSY_MODELS || !psy_models[model].window || !psy_models[model].process){
         av_log(avctx, AV_LOG_ERROR, "Invalid psy model\n");
         return -1;
    }

#ifndef CONFIG_HARDCODED_TABLES
   for (i = 0; i < 316; i++)
        ff_aac_pow2sf_tab[i] = pow(2, (i - 200)/4.);
#endif /* CONFIG_HARDCODED_TABLES */

    ctx->avctx = avctx;
    ctx->flags = flags;
    ctx->bands1024 = bands1024;
    ctx->num_bands1024 = num_bands1024;
    ctx->bands128 = bands128;
    ctx->num_bands128 = num_bands128;
    ctx->model = &psy_models[model];

    if(ctx->flags & PSY_MODEL_NO_ST_ATT || PSY_MODEL_MODE(ctx->flags) == PSY_MODE_QUALITY){
        ctx->flags |= PSY_MODEL_NO_ST_ATT;
        ctx->stereo_att = 0.5f;
    }else{
        ctx->stereo_att = av_clipf(avctx->bit_rate / elements / 192000.0, 0.0f, 0.5f);
    }
    if(ctx->flags & PSY_MODEL_NO_LOWPASS || PSY_MODEL_MODE(ctx->flags) == PSY_MODE_QUALITY){
        ctx->flags |= PSY_MODEL_NO_LOWPASS;
    }else{
        int cutoff;
        cutoff = avctx->bit_rate / elements / 8;
        if(ff_lowpass_filter_init_coeffs(&ctx->lp_coeffs, avctx->sample_rate/2, cutoff) < 0){
            ctx->flags |= PSY_MODEL_NO_LOWPASS;
        }else{
            ctx->lp_state = av_mallocz(sizeof(LPFilterState) * elements * 2);
        }
    }
    if(ctx->model->init)
        return ctx->model->init(ctx, elements);
    return 0;
}

void ff_aac_psy_suggest_window(AACPsyContext *ctx, int16_t *audio, int16_t *la, int tag, int type, ChannelElement *cpe)
{
    ctx->model->window(ctx, audio, la, tag, type, cpe);
}

void ff_aac_psy_analyze(AACPsyContext *ctx, int tag, int type, ChannelElement *cpe)
{
    ctx->model->process(ctx, tag, type, cpe);
}

void av_cold ff_aac_psy_end(AACPsyContext *ctx)
{
    av_freep(&ctx->lp_state);
    if(ctx->model->end)
        return ctx->model->end(ctx);
}

void ff_aac_psy_preprocess(AACPsyContext *ctx, int16_t *audio, int16_t *dest, int tag, int type)
{
    int chans = type == ID_CPE ? 2 : 1;
    const int chstride = ctx->avctx->channels;
    int i, ch;
    float t[2];

    if(chans == 1){
        for(ch = 0; ch < chans; ch++){
            for(i = 0; i < 1024; i++){
                dest[i * chstride + ch] = audio[i * chstride + ch];
            }
        }
    }else{
        for(i = 0; i < 1024; i++){
            if(ctx->flags & PSY_MODEL_NO_ST_ATT){
                for(ch = 0; ch < 2; ch++)
                    t[ch] = audio[i * chstride + ch];
            }else{
                t[0] = audio[i * chstride + 0] * (0.5 + ctx->stereo_att) + audio[i * chstride + 1] * (0.5 - ctx->stereo_att);
                t[1] = audio[i * chstride + 0] * (0.5 - ctx->stereo_att) + audio[i * chstride + 1] * (0.5 + ctx->stereo_att);
            }
            if(!(ctx->flags & PSY_MODEL_NO_LOWPASS)){
                LPFilterState *is = (LPFilterState*)ctx->lp_state + tag*2;
                for(ch = 0; ch < 2; ch++)
                    t[ch] = ff_lowpass_filter(&ctx->lp_coeffs, is + ch, t[ch]);
            }
            for(ch = 0; ch < 2; ch++)
                dest[i * chstride + ch] = av_clip_int16(t[ch]);
        }
    }
}




More information about the ffmpeg-devel mailing list