Merge the AAC encoder from SoC svn. It is still considered experimental.

git-svn-id: file:///var/local/repositories/ffmpeg/trunk@19375 9553f0bf-9b14-0410-a0b8-cfaf0461ba5b

Changelog

@@ -27,6 +27,7 @@ version <next>:
 - Electronic Arts Madcow decoder
 - DivX (XSUB) subtitle encoder
 - nonfree libamr support for AMR-NB/WB decoding/encoding removed
+- Experimental AAC encoder
 
 
 

libavcodec/Makefile

@@ -36,6 +36,7 @@ OBJS-$(CONFIG_VDPAU)                   += vdpau.o
 
 # decoders/encoders/hardware accelerators
 OBJS-$(CONFIG_AAC_DECODER)             += aac.o aactab.o mpeg4audio.o aac_parser.o aac_ac3_parser.o
+OBJS-$(CONFIG_AAC_ENCODER)             += aacenc.o aaccoder.o aacpsy.o aactab.o psymodel.o iirfilter.o mdct.o fft.o mpeg4audio.o
 OBJS-$(CONFIG_AASC_DECODER)            += aasc.o msrledec.o
 OBJS-$(CONFIG_AC3_DECODER)             += eac3dec.o ac3dec.o ac3tab.o ac3dec_data.o ac3.o
 OBJS-$(CONFIG_AC3_ENCODER)             += ac3enc.o ac3tab.o ac3.o

libavcodec/aac.h

@@ -116,6 +116,12 @@ typedef struct {
 
 #define MAX_PREDICTORS 672
 
+#define SCALE_DIV_512    36    ///< scalefactor difference that corresponds to scale difference in 512 times
+#define SCALE_ONE_POS   140    ///< scalefactor index that corresponds to scale=1.0
+#define SCALE_MAX_POS   255    ///< scalefactor index maximum value
+#define SCALE_MAX_DIFF   60    ///< maximum scalefactor difference allowed by standard
+#define SCALE_DIFF_ZERO  60    ///< codebook index corresponding to zero scalefactor indices difference
+
 /**
  * Individual Channel Stream
  */
@@ -126,6 +132,7 @@ typedef struct {
     int num_window_groups;
     uint8_t group_len[8];
     const uint16_t *swb_offset; ///< table of offsets to the lowest spectral coefficient of a scalefactor band, sfb, for a particular window
+    const uint8_t *swb_sizes;   ///< table of scalefactor band sizes for a particular window
     int num_swb;                ///< number of scalefactor window bands
     int num_windows;
     int tns_max_bands;
@@ -165,6 +172,7 @@ typedef struct {
 
 typedef struct {
     int num_pulse;
+    int start;
     int pos[4];
     int amp[4];
 } Pulse;
@@ -189,11 +197,14 @@ typedef struct {
 typedef struct {
     IndividualChannelStream ics;
     TemporalNoiseShaping tns;
-    enum BandType band_type[120];             ///< band types
+    Pulse pulse;
+    enum BandType band_type[128];             ///< band types
     int band_type_run_end[120];               ///< band type run end points
     float sf[120];                            ///< scalefactors
+    int sf_idx[128];                          ///< scalefactor indices (used by encoder)
+    uint8_t zeroes[128];                      ///< band is not coded (used by encoder)
     DECLARE_ALIGNED_16(float, coeffs[1024]);  ///< coefficients for IMDCT
-    DECLARE_ALIGNED_16(float, saved[512]);    ///< overlap
+    DECLARE_ALIGNED_16(float, saved[1024]);   ///< overlap
     DECLARE_ALIGNED_16(float, ret[1024]);     ///< PCM output
     PredictorState predictor_state[MAX_PREDICTORS];
 } SingleChannelElement;
@@ -203,7 +214,9 @@ typedef struct {
  */
 typedef struct {
     // CPE specific
-    uint8_t ms_mask[120];     ///< Set if mid/side stereo is used for each scalefactor window band
+    int common_window;        ///< Set if channels share a common 'IndividualChannelStream' in bitstream.
+    int     ms_mode;          ///< Signals mid/side stereo flags coding mode (used by encoder)
+    uint8_t ms_mask[128];     ///< Set if mid/side stereo is used for each scalefactor window band
     // shared
     SingleChannelElement ch[2];
     // CCE specific

libavcodec/aacenc.h

+/*
+ * AAC encoder
+ * 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 AVCODEC_AACENC_H
+#define AVCODEC_AACENC_H
+
+#include "avcodec.h"
+#include "put_bits.h"
+#include "dsputil.h"
+
+#include "aac.h"
+
+#include "psymodel.h"
+
+struct AACEncContext;
+
+typedef struct AACCoefficientsEncoder{
+    void (*search_for_quantizers)(AVCodecContext *avctx, struct AACEncContext *s,
+                                  SingleChannelElement *sce, const float lambda);
+    void (*encode_window_bands_info)(struct AACEncContext *s, SingleChannelElement *sce,
+                                     int win, int group_len, const float lambda);
+    void (*quantize_and_encode_band)(struct AACEncContext *s, PutBitContext *pb, const float *in, int size,
+                                     int scale_idx, int cb, const float lambda);
+    void (*search_for_ms)(struct AACEncContext *s, ChannelElement *cpe, const float lambda);
+}AACCoefficientsEncoder;
+
+extern AACCoefficientsEncoder ff_aac_coders[];
+
+/**
+ * AAC encoder context
+ */
+typedef struct AACEncContext {
+    PutBitContext pb;
+    MDCTContext mdct1024;                        ///< long (1024 samples) frame transform context
+    MDCTContext mdct128;                         ///< short (128 samples) frame transform context
+    DSPContext  dsp;
+    DECLARE_ALIGNED_16(FFTSample, output[2048]); ///< temporary buffer for MDCT input coefficients
+    int16_t* samples;                            ///< saved preprocessed input
+
+    int samplerate_index;                        ///< MPEG-4 samplerate index
+
+    ChannelElement *cpe;                         ///< channel elements
+    FFPsyContext psy;
+    struct FFPsyPreprocessContext* psypp;
+    AACCoefficientsEncoder *coder;
+    int cur_channel;
+    int last_frame;
+    float lambda;
+    DECLARE_ALIGNED_16(int,   qcoefs[96][2]);    ///< quantized coefficients
+    DECLARE_ALIGNED_16(float, scoefs[1024]);     ///< scaled coefficients
+} AACEncContext;
+
+#endif /* AVCODEC_AACENC_H */

libavcodec/allcodecs.c

@@ -195,7 +195,7 @@ void avcodec_register_all(void)
     REGISTER_ENCDEC  (ZMBV, zmbv);
 
     /* audio codecs */
-    REGISTER_DECODER (AAC, aac);
+    REGISTER_ENCDEC  (AAC, aac);
     REGISTER_ENCDEC  (AC3, ac3);
     REGISTER_ENCDEC  (ALAC, alac);
     REGISTER_DECODER (APE, ape);

libavcodec/psymodel.c

+/*
+ * audio 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
+ */
+
+#include "avcodec.h"
+#include "psymodel.h"
+#include "iirfilter.h"
+
+extern const FFPsyModel ff_aac_psy_model;
+
+av_cold int ff_psy_init(FFPsyContext *ctx, AVCodecContext *avctx,
+                        int num_lens,
+                        const uint8_t **bands, const int* num_bands)
+{
+    ctx->avctx = avctx;
+    ctx->psy_bands = av_mallocz(sizeof(FFPsyBand) * PSY_MAX_BANDS * avctx->channels);
+    ctx->bands     = av_malloc (sizeof(ctx->bands[0])     * num_lens);
+    ctx->num_bands = av_malloc (sizeof(ctx->num_bands[0]) * num_lens);
+    memcpy(ctx->bands,     bands,     sizeof(ctx->bands[0])     *  num_lens);
+    memcpy(ctx->num_bands, num_bands, sizeof(ctx->num_bands[0]) *  num_lens);
+    switch(ctx->avctx->codec_id){
+    case CODEC_ID_AAC:
+        ctx->model = &ff_aac_psy_model;
+        break;
+    }
+    if(ctx->model->init)
+        return ctx->model->init(ctx);
+    return 0;
+}
+
+FFPsyWindowInfo ff_psy_suggest_window(FFPsyContext *ctx,
+                                      const int16_t *audio, const int16_t *la,
+                                      int channel, int prev_type)
+{
+    return ctx->model->window(ctx, audio, la, channel, prev_type);
+}
+
+void ff_psy_set_band_info(FFPsyContext *ctx, int channel,
+                          const float *coeffs, FFPsyWindowInfo *wi)
+{
+    ctx->model->analyze(ctx, channel, coeffs, wi);
+}
+
+av_cold void ff_psy_end(FFPsyContext *ctx)
+{
+    if(ctx->model->end)
+        ctx->model->end(ctx);
+    av_freep(&ctx->bands);
+    av_freep(&ctx->num_bands);
+    av_freep(&ctx->psy_bands);
+}
+
+typedef struct FFPsyPreprocessContext{
+    AVCodecContext *avctx;
+    float stereo_att;
+    struct FFIIRFilterCoeffs *fcoeffs;
+    struct FFIIRFilterState **fstate;
+}FFPsyPreprocessContext;
+
+#define FILT_ORDER 4
+
+av_cold struct FFPsyPreprocessContext* ff_psy_preprocess_init(AVCodecContext *avctx)
+{
+    FFPsyPreprocessContext *ctx;
+    int i;
+    float cutoff_coeff;
+    ctx = av_mallocz(sizeof(FFPsyPreprocessContext));
+    ctx->avctx = avctx;
+
+    if(avctx->flags & CODEC_FLAG_QSCALE)
+        cutoff_coeff = 1.0f / av_clip(1 + avctx->global_quality / FF_QUALITY_SCALE, 1, 8);
+    else
+        cutoff_coeff = avctx->bit_rate / (4.0f * avctx->sample_rate * avctx->channels);
+
+    ctx->fcoeffs = ff_iir_filter_init_coeffs(FF_FILTER_TYPE_BUTTERWORTH, FF_FILTER_MODE_LOWPASS,
+                                           FILT_ORDER, cutoff_coeff, 0.0, 0.0);
+    if(ctx->fcoeffs){
+        ctx->fstate = av_mallocz(sizeof(ctx->fstate[0]) * avctx->channels);
+        for(i = 0; i < avctx->channels; i++)
+            ctx->fstate[i] = ff_iir_filter_init_state(FILT_ORDER);
+    }
+    return ctx;
+}
+
+void ff_psy_preprocess(struct FFPsyPreprocessContext *ctx,
+                       const int16_t *audio, int16_t *dest,
+                       int tag, int channels)
+{
+    int ch, i;
+    if(ctx->fstate){
+        for(ch = 0; ch < channels; ch++){
+            ff_iir_filter(ctx->fcoeffs, ctx->fstate[tag+ch], ctx->avctx->frame_size,
+                          audio + ch, ctx->avctx->channels,
+                          dest  + ch, ctx->avctx->channels);
+        }
+    }else{
+        for(ch = 0; ch < channels; ch++){
+            for(i = 0; i < ctx->avctx->frame_size; i++)
+                dest[i*ctx->avctx->channels + ch] = audio[i*ctx->avctx->channels + ch];
+        }
+    }
+}
+
+av_cold void ff_psy_preprocess_end(struct FFPsyPreprocessContext *ctx)
+{
+    int i;
+    ff_iir_filter_free_coeffs(ctx->fcoeffs);
+    if (ctx->fstate)
+        for (i = 0; i < ctx->avctx->channels; i++)
+            ff_iir_filter_free_state(ctx->fstate[i]);
+    av_freep(&ctx->fstate);
+}
+

libavcodec/psymodel.h

+/*
+ * audio 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 AVCODEC_PSYMODEL_H
+#define AVCODEC_PSYMODEL_H
+
+#include "avcodec.h"
+
+/** maximum possible number of bands */
+#define PSY_MAX_BANDS 128
+
+/**
+ * single band psychoacoustic information
+ */
+typedef struct FFPsyBand{
+    int   bits;
+    float energy;
+    float threshold;
+    float distortion;
+    float perceptual_weight;
+}FFPsyBand;
+
+/**
+ * windowing related information
+ */
+typedef struct FFPsyWindowInfo{
+    int window_type[3];               ///< window type (short/long/transitional, etc.) - current, previous and next
+    int window_shape;                 ///< window shape (sine/KBD/whatever)
+    int num_windows;                  ///< number of windows in a frame
+    int grouping[8];                  ///< window grouping (for e.g. AAC)
+    int *window_sizes;                ///< sequence of window sizes inside one frame (for eg. WMA)
+}FFPsyWindowInfo;
+
+/**
+ * context used by psychoacoustic model
+ */
+typedef struct FFPsyContext{
+    AVCodecContext *avctx;            ///< encoder context
+    const struct FFPsyModel *model;   ///< encoder-specific model functions
+
+    FFPsyBand *psy_bands;             ///< frame bands information
+
+    uint8_t **bands;                  ///< scalefactor band sizes for possible frame sizes
+    int     *num_bands;               ///< number of scalefactor bands for possible frame sizes
+    int num_lens;                     ///< number of scalefactor band sets
+
+    void* model_priv_data;            ///< psychoacoustic model implementation private data
+}FFPsyContext;
+
+/**
+ * codec-specific psychoacoustic model implementation
+ */
+typedef struct FFPsyModel {
+    const char *name;
+    int  (*init)   (FFPsyContext *apc);
+    FFPsyWindowInfo (*window)(FFPsyContext *ctx, const int16_t *audio, const int16_t *la, int channel, int prev_type);
+    void (*analyze)(FFPsyContext *ctx, int channel, const float *coeffs, FFPsyWindowInfo *wi);
+    void (*end)    (FFPsyContext *apc);
+}FFPsyModel;
+
+/**
+ * Initialize psychoacoustic model.
+ *
+ * @param ctx        model context
+ * @param avctx      codec context
+ * @param num_lens   number of possible frame lengths
+ * @param bands      scalefactor band lengths for all frame lengths
+ * @param num_bands  number of scalefactor bands for all frame lengths
+ *
+ * @return zero if successful, a negative value if not
+ */
+av_cold int ff_psy_init(FFPsyContext *ctx, AVCodecContext *avctx,
+                        int num_lens,
+                        const uint8_t **bands, const int* num_bands);
+
+/**
+ * Suggest window sequence for channel.
+ *
+ * @param ctx       model context
+ * @param audio     samples for the current frame
+ * @param la        lookahead samples (NULL when unavailable)
+ * @param channel   number of channel element to analyze
+ * @param prev_type previous window type
+ *
+ * @return suggested window information in a structure
+ */
+FFPsyWindowInfo ff_psy_suggest_window(FFPsyContext *ctx,
+                                      const int16_t *audio, const int16_t *la,
+                                      int channel, int prev_type);
+
+
+/**
+ * Perform psychoacoustic analysis and set band info (threshold, energy).
+ *
+ * @param ctx     model context
+ * @param channel audio channel number
+ * @param coeffs  pointer to the transformed coefficients
+ * @param wi      window information
+ */
+void ff_psy_set_band_info(FFPsyContext *ctx, int channel, const float *coeffs,
+                          FFPsyWindowInfo *wi);
+
+/**
+ * Cleanup model context at the end.
+ *
+ * @param ctx model context
+ */
+av_cold void ff_psy_end(FFPsyContext *ctx);
+
+
+/**************************************************************************
+ *                       Audio preprocessing stuff.                       *
+ *       This should be moved into some audio filter eventually.          *
+ **************************************************************************/
+struct FFPsyPreprocessContext;
+
+/**
+ * psychoacoustic model audio preprocessing initialization
+ */
+av_cold struct FFPsyPreprocessContext* ff_psy_preprocess_init(AVCodecContext *avctx);
+
+/**
+ * Preprocess several channel in audio frame in order to compress it better.
+ *
+ * @param ctx      preprocessing context
+ * @param audio    samples to preprocess
+ * @param dest     place to put filtered samples
+ * @param tag      channel number
+ * @param channels number of channel to preprocess (some additional work may be done on stereo pair)
+ */
+void ff_psy_preprocess(struct FFPsyPreprocessContext *ctx,
+                       const int16_t *audio, int16_t *dest,
+                       int tag, int channels);
+
+/**
+ * Cleanup audio preprocessing module.
+ */
+av_cold void ff_psy_preprocess_end(struct FFPsyPreprocessContext *ctx);
+
+#endif /* AVCODEC_PSYMODEL_H */