SIPR16k decoder

Originally committed as revision 21234 to svn://svn.ffmpeg.org/ffmpeg/trunk

Changelog

@@ -48,7 +48,7 @@ version <next>:
 - R210 decoder
 - Auravision Aura 1 and 2 decoders
 - Deluxe Paint Animation playback system
-- SIPR decoding for modes 8k5, 6k5 and 5k0
+- SIPR decoder
 - Adobe Filmstrip muxer and demuxer
 
 

doc/general.texi

@@ -620,7 +620,6 @@ following image formats are supported:
 @item RealAudio 3.0 (dnet)   @tab IX  @tab  X
     @tab Real low bitrate AC-3 codec
 @item RealAudio SIPR / ACELP.NET @tab     @tab  X
-    @tab 16 kbps mode not yet supported
 @item Shorten                @tab     @tab  X
 @item Sierra VMD audio       @tab     @tab  X
     @tab Used in Sierra VMD files.

libavcodec/Makefile

@@ -263,7 +263,8 @@ OBJS-$(CONFIG_SGI_ENCODER)             += sgienc.o rle.o
 OBJS-$(CONFIG_SHORTEN_DECODER)         += shorten.o
 OBJS-$(CONFIG_SIPR_DECODER)            += sipr.o acelp_pitch_delay.o \
                                           celp_math.o acelp_vectors.o \
-                                          acelp_filters.o celp_filters.o lsp.o
+                                          acelp_filters.o celp_filters.o lsp.o \
+                                          sipr16k.o
 OBJS-$(CONFIG_SMACKAUD_DECODER)        += smacker.o
 OBJS-$(CONFIG_SMACKER_DECODER)         += smacker.o
 OBJS-$(CONFIG_SMC_DECODER)             += smc.o

libavcodec/avcodec.h

@@ -30,7 +30,7 @@
 #include "libavutil/avutil.h"
 
 #define LIBAVCODEC_VERSION_MAJOR 52
-#define LIBAVCODEC_VERSION_MINOR 47
+#define LIBAVCODEC_VERSION_MINOR 48
 #define LIBAVCODEC_VERSION_MICRO  0
 
 #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \

libavcodec/sipr.c

@@ -51,6 +51,7 @@ typedef struct {
 
     /* bitstream parameters */
     uint8_t number_of_fc_indexes;
+    uint8_t ma_predictor_bits;  ///< size in bits of the switched MA predictor
 
     /** size in bits of the i-th stage vector of quantizer */
     uint8_t vq_indexes_bits[5];
@@ -64,6 +65,22 @@ typedef struct {
 } SiprModeParam;
 
 static const SiprModeParam modes[MODE_COUNT] = {
+    [MODE_16k] = {
+        .mode_name          = "16k",
+        .bits_per_frame     = 160,
+        .subframe_count     = SUBFRAME_COUNT_16k,
+        .frames_per_packet  = 1,
+        .pitch_sharp_factor = 0.00,
+
+        .number_of_fc_indexes = 10,
+        .ma_predictor_bits    = 1,
+        .vq_indexes_bits      = {7, 8, 7, 7, 7},
+        .pitch_delay_bits     = {9, 6},
+        .gp_index_bits        = 4,
+        .fc_index_bits        = {4, 5, 4, 5, 4, 5, 4, 5, 4, 5},
+        .gc_index_bits        = 5
+    },
+
     [MODE_8k5] = {
         .mode_name          = "8k5",
         .bits_per_frame     = 152,
@@ -72,6 +89,7 @@ static const SiprModeParam modes[MODE_COUNT] = {
         .pitch_sharp_factor = 0.8,
 
         .number_of_fc_indexes = 3,
+        .ma_predictor_bits    = 0,
         .vq_indexes_bits      = {6, 7, 7, 7, 5},
         .pitch_delay_bits     = {8, 5, 5},
         .gp_index_bits        = 0,
@@ -87,6 +105,7 @@ static const SiprModeParam modes[MODE_COUNT] = {
         .pitch_sharp_factor = 0.8,
 
         .number_of_fc_indexes = 3,
+        .ma_predictor_bits    = 0,
         .vq_indexes_bits      = {6, 7, 7, 7, 5},
         .pitch_delay_bits     = {8, 5, 5},
         .gp_index_bits        = 0,
@@ -102,6 +121,7 @@ static const SiprModeParam modes[MODE_COUNT] = {
         .pitch_sharp_factor = 0.85,
 
         .number_of_fc_indexes = 1,
+        .ma_predictor_bits    = 0,
         .vq_indexes_bits      = {6, 7, 7, 7, 5},
         .pitch_delay_bits     = {8, 5, 8, 5, 5},
         .gp_index_bits        = 0,
@@ -173,6 +193,8 @@ static void decode_parameters(SiprParameters* parms, GetBitContext *pgb,
 {
     int i, j;
 
+    parms->ma_pred_switch           = get_bits(pgb, p->ma_predictor_bits);
+
     for (i = 0; i < 5; i++)
         parms->vq_indexes[i]        = get_bits(pgb, p->vq_indexes_bits[i]);
 
@@ -490,6 +512,9 @@ static av_cold int sipr_decoder_init(AVCodecContext * avctx)
 
     av_log(avctx, AV_LOG_DEBUG, "Mode: %s\n", modes[ctx->mode].mode_name);
 
+    if (ctx->mode == MODE_16k)
+        ff_sipr_init_16k(ctx);
+
     for (i = 0; i < LP_FILTER_ORDER; i++)
         ctx->lsp_history[i] = cos((i+1) * M_PI / (LP_FILTER_ORDER + 1));
 
@@ -498,12 +523,6 @@ static av_cold int sipr_decoder_init(AVCodecContext * avctx)
 
     avctx->sample_fmt = SAMPLE_FMT_FLT;
 
-    if (ctx->mode == MODE_16k) {
-        av_log(avctx, AV_LOG_ERROR, "decoding 16kbps SIPR files is not "
-                                    "supported yet.\n");
-        return -1;
-    }
-
     dsputil_init(&ctx->dsp, avctx);
 
     return 0;
@@ -518,6 +537,7 @@ static int sipr_decode_frame(AVCodecContext *avctx, void *datap,
     const SiprModeParam *mode_par = &modes[ctx->mode];
     GetBitContext gb;
     float *data = datap;
+    int subframe_size = ctx->mode == MODE_16k ? L_SUBFR_16k : SUBFR_SIZE;
     int i;
 
     ctx->avctx = avctx;
@@ -529,7 +549,7 @@ static int sipr_decode_frame(AVCodecContext *avctx, void *datap,
         *data_size = 0;
         return -1;
     }
-    if (*data_size < SUBFR_SIZE * mode_par->subframe_count * sizeof(float)) {
+    if (*data_size < subframe_size * mode_par->subframe_count * sizeof(float)) {
         av_log(avctx, AV_LOG_ERROR,
                "Error processing packet: output buffer (%d) too small\n",
                *data_size);
@@ -542,12 +562,16 @@ static int sipr_decode_frame(AVCodecContext *avctx, void *datap,
 
     for (i = 0; i < mode_par->frames_per_packet; i++) {
         decode_parameters(&parm, &gb, mode_par);
-        decode_frame(ctx, &parm, data);
 
-        data += SUBFR_SIZE * mode_par->subframe_count;
+        if (ctx->mode == MODE_16k)
+            ff_sipr_decode_frame_16k(ctx, &parm, data);
+        else
+            decode_frame(ctx, &parm, data);
+
+        data += subframe_size * mode_par->subframe_count;
     }
 
-    *data_size = mode_par->frames_per_packet * SUBFR_SIZE *
+    *data_size = mode_par->frames_per_packet * subframe_size *
         mode_par->subframe_count * sizeof(float);
 
     return mode_par->bits_per_frame >> 3;

libavcodec/sipr.h

@@ -28,6 +28,11 @@
 #include "dsputil.h"
 #include "acelp_pitch_delay.h"
 
+#define LP_FILTER_ORDER_16k  16
+#define L_SUBFR_16k          80
+#define PITCH_MIN            30
+#define PITCH_MAX            281
+
 #define LSFQ_DIFF_MIN        (0.0125 * M_PI)
 
 #define LP_FILTER_ORDER      10
@@ -38,6 +43,8 @@
 /**  Subframe size for all modes except 16k */
 #define SUBFR_SIZE           48
 
+#define SUBFRAME_COUNT_16k   2
+
 typedef enum {
     MODE_16k,
     MODE_8k5,
@@ -53,9 +60,9 @@ typedef struct {
     SiprMode mode;
 
     float past_pitch_gain;
-    float lsf_history[LP_FILTER_ORDER];
+    float lsf_history[LP_FILTER_ORDER_16k];
 
-    float excitation[L_INTERPOL + PITCH_DELAY_MAX + 5*SUBFR_SIZE];
+    float excitation[L_INTERPOL + PITCH_MAX + 2 * L_SUBFR_16k];
 
     DECLARE_ALIGNED_16(float, synth_buf[LP_FILTER_ORDER + 5*SUBFR_SIZE + 6]);
 
@@ -70,9 +77,19 @@ typedef struct {
     float postfilter_agc;
     float postfilter_mem5k0[PITCH_DELAY_MAX + LP_FILTER_ORDER];
     float postfilter_syn5k0[LP_FILTER_ORDER + SUBFR_SIZE*5];
+
+    /* 16k */
+    int pitch_lag_prev;
+    float iir_mem[LP_FILTER_ORDER_16k+1];
+    float filt_buf[2][LP_FILTER_ORDER_16k+1];
+    float *filt_mem[2];
+    float mem_preemph[LP_FILTER_ORDER_16k];
+    float synth[LP_FILTER_ORDER_16k];
+    double lsp_history_16k[16];
 } SiprContext;
 
 typedef struct {
+    int ma_pred_switch;        ///< switched moving average predictor
     int vq_indexes[5];
     int pitch_delay[5];        ///< pitch delay
     int gp_index[5];           ///< adaptive-codebook gain indexes
@@ -80,4 +97,11 @@ typedef struct {
     int gc_index[5];           ///< fixed-codebook gain indexes
 } SiprParameters;
 
+extern const float ff_pow_0_5[16];
+
+void ff_sipr_init_16k(SiprContext *ctx);
+
+void ff_sipr_decode_frame_16k(SiprContext *ctx, SiprParameters *params,
+                              float *out_data);
+
 #endif /* AVCODEC_SIPR_H */

libavcodec/sipr16k.c

+/*
+ * SIPR decoder for the 16k mode
+ *
+ * Copyright (c) 2008 Vladimir Voroshilov
+ * Copyright (c) 2009 Vitor Sessak
+ *
+ * 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 <math.h>
+
+#include "sipr.h"
+#include "libavutil/mathematics.h"
+#include "lsp.h"
+#include "celp_math.h"
+#include "acelp_vectors.h"
+#include "acelp_pitch_delay.h"
+#include "acelp_filters.h"
+#include "celp_filters.h"
+
+#include "sipr16kdata.h"
+
+/**
+ * Convert an lsf vector into an lsp vector.
+ *
+ * @param lsf               input lsf vector
+ * @param lsp               output lsp vector
+ */
+static void lsf2lsp(const float *lsf, double *lsp)
+{
+    int i;
+
+    for (i = 0; i < LP_FILTER_ORDER_16k; i++)
+        lsp[i] = cosf(lsf[i]);
+}
+
+static void dequant(float *out, const int *idx, const float *cbs[])
+{
+    int i;
+
+    for (i = 0; i < 4; i++)
+        memcpy(out + 3*i, cbs[i] + 3*idx[i], 3*sizeof(float));
+
+    memcpy(out + 12, cbs[4] + 4*idx[4], 4*sizeof(float));
+}
+
+static void lsf_decode_fp_16k(float* lsf_history, float* isp_new,
+                              const int* parm, int ma_pred)
+{
+    int i;
+    float isp_q[LP_FILTER_ORDER_16k];
+
+    dequant(isp_q, parm, lsf_codebooks_16k);
+
+    for (i = 0; i < LP_FILTER_ORDER_16k; i++) {
+        isp_new[i] = (1 - qu[ma_pred]) * isp_q[i]
+                    +     qu[ma_pred]  * lsf_history[i]
+                    + mean_lsf_16k[i];
+    }
+
+    memcpy(lsf_history, isp_q, LP_FILTER_ORDER_16k * sizeof(float));
+}
+
+static int dec_delay3_1st(int index)
+{
+    if (index < 390) {
+        return index + 88;
+    } else
+        return 3 * index - 690;
+}
+
+static int dec_delay3_2nd(int index, int pit_min, int pit_max,
+                          int pitch_lag_prev)
+{
+    if (index < 62) {
+        int pitch_delay_min = av_clip(pitch_lag_prev - 10,
+                                      pit_min, pit_max - 19);
+        return 3 * pitch_delay_min + index - 2;
+    } else
+        return 3 * pitch_lag_prev;
+}
+
+static void postfilter(float* synth, float* iir_mem, float* filt_mem[2],
+                       float* mem_preemph)
+{
+    float buf[30 + LP_FILTER_ORDER_16k];
+    float *tmpbuf = buf + LP_FILTER_ORDER_16k;
+    float s;
+    int i;
+
+    for (i = 0; i < LP_FILTER_ORDER_16k; i++)
+        filt_mem[0][i] = iir_mem[i] * ff_pow_0_5[i];
+
+    memcpy(tmpbuf - LP_FILTER_ORDER_16k, mem_preemph,
+           LP_FILTER_ORDER_16k*sizeof(*buf));
+
+    ff_celp_lp_synthesis_filterf(tmpbuf, filt_mem[1], synth, 30,
+                                 LP_FILTER_ORDER_16k);
+
+    memcpy(synth - LP_FILTER_ORDER_16k, mem_preemph,
+           LP_FILTER_ORDER_16k * sizeof(*synth));
+
+    ff_celp_lp_synthesis_filterf(synth, filt_mem[0], synth, 2*L_SUBFR_16k,
+                                 LP_FILTER_ORDER_16k);
+
+    memcpy(mem_preemph, synth + 2*L_SUBFR_16k - LP_FILTER_ORDER_16k,
+           LP_FILTER_ORDER_16k * sizeof(*synth));
+
+    FFSWAP(float *, filt_mem[0], filt_mem[1]);
+    for (i = 0, s = 0; i < 30; i++, s += 1.0/30)
+        synth[i] = tmpbuf[i] + s * (synth[i] - tmpbuf[i]);
+}
+
+/**
+ * Floating point version of ff_acelp_lp_decode().
+ */
+static void acelp_lp_decodef(float *lp_1st, float *lp_2nd,
+                             const double *lsp_2nd, const double *lsp_prev)
+{
+    double lsp_1st[LP_FILTER_ORDER_16k];
+    int i;
+
+    /* LSP values for first subframe (3.2.5 of G.729, Equation 24) */
+    for (i = 0; i < LP_FILTER_ORDER_16k; i++)
+        lsp_1st[i] = (lsp_2nd[i] + lsp_prev[i]) * 0.5;
+
+    ff_acelp_lspd2lpc(lsp_1st, lp_1st, LP_FILTER_ORDER_16k >> 1);
+
+    /* LSP values for second subframe (3.2.5 of G.729) */
+    ff_acelp_lspd2lpc(lsp_2nd, lp_2nd, LP_FILTER_ORDER_16k >> 1);
+}
+
+/**
+ * Floating point version of ff_acelp_decode_gain_code().
+ */
+static float acelp_decode_gain_codef(float gain_corr_factor, const float *fc_v,
+                                     float mr_energy, const float *quant_energy,
+                                     const float *ma_prediction_coeff,
+                                     int subframe_size, int ma_pred_order)
+{
+    mr_energy +=
+        ff_dot_productf(quant_energy, ma_prediction_coeff, ma_pred_order);
+
+    mr_energy = gain_corr_factor * exp(M_LN10 / 20. * mr_energy) /
+        sqrt((0.01 + ff_dot_productf(fc_v, fc_v, subframe_size)));
+    return mr_energy;
+}
+
+#define DIVIDE_BY_3(x) ((x) * 10923 >> 15)
+
+void ff_sipr_decode_frame_16k(SiprContext *ctx, SiprParameters *params,
+                              float *out_data)
+{
+    int frame_size = SUBFRAME_COUNT_16k * L_SUBFR_16k;
+    float *synth = ctx->synth_buf + LP_FILTER_ORDER_16k;
+    float lsf_new[LP_FILTER_ORDER_16k];
+    double lsp_new[LP_FILTER_ORDER_16k];
+    float Az[2][LP_FILTER_ORDER_16k];
+    float fixed_vector[L_SUBFR_16k];
+    float pitch_fac, gain_code;
+
+    int i;
+    int pitch_delay_3x;
+
+    float *excitation = ctx->excitation + 292;
+
+    lsf_decode_fp_16k(ctx->lsf_history, lsf_new, params->vq_indexes,
+                      params->ma_pred_switch);
+
+    ff_set_min_dist_lsf(lsf_new, LSFQ_DIFF_MIN / 2, LP_FILTER_ORDER_16k);
+
+    lsf2lsp(lsf_new, lsp_new);
+
+    acelp_lp_decodef(Az[0], Az[1], lsp_new, ctx->lsp_history_16k);
+
+    memcpy(ctx->lsp_history_16k, lsp_new, LP_FILTER_ORDER_16k * sizeof(double));
+
+    memcpy(synth - LP_FILTER_ORDER_16k, ctx->synth,
+           LP_FILTER_ORDER_16k * sizeof(*synth));
+
+    for (i = 0; i < SUBFRAME_COUNT_16k; i++) {
+        int i_subfr = i * L_SUBFR_16k;
+        AMRFixed f;
+        float gain_corr_factor;
+        int pitch_delay_int;
+        int pitch_delay_frac;
+
+        if (!i) {
+            pitch_delay_3x = dec_delay3_1st(params->pitch_delay[i]);
+        } else
+            pitch_delay_3x = dec_delay3_2nd(params->pitch_delay[i],
+                                            PITCH_MIN, PITCH_MAX,
+                                            ctx->pitch_lag_prev);
+
+        pitch_fac = gain_pitch_cb_16k[params->gp_index[i]];
+        f.pitch_fac = FFMIN(pitch_fac, 1.0);
+        f.pitch_lag = DIVIDE_BY_3(pitch_delay_3x+1);
+        ctx->pitch_lag_prev = f.pitch_lag;
+
+        pitch_delay_int  = DIVIDE_BY_3(pitch_delay_3x + 2);
+        pitch_delay_frac = pitch_delay_3x + 2 - 3*pitch_delay_int;
+
+        ff_acelp_interpolatef(&excitation[i_subfr],
+                              &excitation[i_subfr] - pitch_delay_int + 1,
+                              sinc_win, 3, pitch_delay_frac + 1,
+                              LP_FILTER_ORDER, L_SUBFR_16k);
+
+
+        memset(fixed_vector, 0, sizeof(fixed_vector));
+
+        ff_decode_10_pulses_35bits(params->fc_indexes[i], &f,
+                                   ff_fc_4pulses_8bits_tracks_13, 5, 4);
+
+        ff_set_fixed_vector(fixed_vector, &f, 1.0, L_SUBFR_16k);
+
+        gain_corr_factor = gain_cb_16k[params->gc_index[i]];
+        gain_code = gain_corr_factor *
+            acelp_decode_gain_codef(sqrt(L_SUBFR_16k), fixed_vector,
+                                    19.0 - 15.0/(0.05*M_LN10/M_LN2),
+                                    pred_16k, ctx->energy_history,
+                                    L_SUBFR_16k, 2);
+
+        ctx->energy_history[1] = ctx->energy_history[0];
+        ctx->energy_history[0] = 20.0 * log10f(gain_corr_factor);
+
+        ff_weighted_vector_sumf(&excitation[i_subfr], &excitation[i_subfr],
+                                fixed_vector, pitch_fac,
+                                gain_code, L_SUBFR_16k);
+
+        ff_celp_lp_synthesis_filterf(synth + i_subfr, Az[i],
+                                     &excitation[i_subfr], L_SUBFR_16k,
+                                     LP_FILTER_ORDER_16k);
+
+    }
+    memcpy(ctx->synth, synth + frame_size - LP_FILTER_ORDER_16k,
+           LP_FILTER_ORDER_16k * sizeof(*synth));
+
+    memmove(ctx->excitation, ctx->excitation + 2 * L_SUBFR_16k,
+            (L_INTERPOL+PITCH_MAX) * sizeof(float));
+
+    postfilter(synth, ctx->iir_mem, ctx->filt_mem, ctx->mem_preemph);
+
+    memcpy(ctx->iir_mem, Az[1], LP_FILTER_ORDER_16k * sizeof(float));
+
+    ctx->dsp.vector_clipf(out_data, synth, -1, 32767./(1<<15), frame_size);
+
+}
+
+void ff_sipr_init_16k(SiprContext *ctx)
+{
+    int i;
+
+    for (i = 0; i < LP_FILTER_ORDER_16k; i++)
+        ctx->lsp_history_16k[i] = cos((i + 1) * M_PI/(LP_FILTER_ORDER_16k + 1));
+
+    ctx->filt_mem[0] = ctx->filt_buf[0];
+    ctx->filt_mem[1] = ctx->filt_buf[1];
+
+    ctx->pitch_lag_prev = 180;
+}