* RealAudio 2.0 (28.8K)
* Copyright (c) 2003 the ffmpeg project
*
- * This file is part of FFmpeg.
+ * This file is part of Libav.
*
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav 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,
+ * Libav 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
+ * License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
+#include "libavutil/float_dsp.h"
#include "avcodec.h"
-#define ALT_BITSTREAM_READER_LE
-#include "bitstream.h"
+#define BITSTREAM_READER_LE
+#include "get_bits.h"
#include "ra288.h"
+#include "lpc.h"
+#include "celp_math.h"
+#include "celp_filters.h"
+
+#define MAX_BACKWARD_FILTER_ORDER 36
+#define MAX_BACKWARD_FILTER_LEN 40
+#define MAX_BACKWARD_FILTER_NONREC 35
+
+#define RA288_BLOCK_SIZE 5
+#define RA288_BLOCKS_PER_FRAME 32
typedef struct {
- float history[8];
- float output[40];
- float pr1[36];
- float pr2[10];
- int phase;
-
- float st1a[111], st1b[37], st1[37];
- float st2a[38], st2b[11], st2[11];
- float sb[41];
- float lhist[10];
-} Real288_internal;
-
-static inline float scalar_product_float(const float * v1, const float * v2,
- int size)
+ AVFrame frame;
+ DSPContext dsp;
+ AVFloatDSPContext fdsp;
+ DECLARE_ALIGNED(32, float, sp_lpc)[FFALIGN(36, 16)]; ///< LPC coefficients for speech data (spec: A)
+ DECLARE_ALIGNED(32, float, gain_lpc)[FFALIGN(10, 16)]; ///< LPC coefficients for gain (spec: GB)
+
+ /** speech data history (spec: SB).
+ * Its first 70 coefficients are updated only at backward filtering.
+ */
+ float sp_hist[111];
+
+ /// speech part of the gain autocorrelation (spec: REXP)
+ float sp_rec[37];
+
+ /** log-gain history (spec: SBLG).
+ * Its first 28 coefficients are updated only at backward filtering.
+ */
+ float gain_hist[38];
+
+ /// recursive part of the gain autocorrelation (spec: REXPLG)
+ float gain_rec[11];
+} RA288Context;
+
+static av_cold int ra288_decode_init(AVCodecContext *avctx)
{
- float res = 0.;
+ RA288Context *ractx = avctx->priv_data;
+ avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
+ avpriv_float_dsp_init(&ractx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
- while (size--)
- res += *v1++ * *v2++;
+ avcodec_get_frame_defaults(&ractx->frame);
+ avctx->coded_frame = &ractx->frame;
- return res;
+ return 0;
}
-/* Decode and produce output */
-static void decode(Real288_internal *glob, float gain, int cb_coef)
+static void convolve(float *tgt, const float *src, int len, int n)
{
- int x, y;
+ for (; n >= 0; n--)
+ tgt[n] = ff_dot_productf(src, src - n, len);
+
+}
+
+static void decode(RA288Context *ractx, float gain, int cb_coef)
+{
+ int i;
double sumsum;
float sum, buffer[5];
+ float *block = ractx->sp_hist + 70 + 36; // current block
+ float *gain_block = ractx->gain_hist + 28;
- memmove(glob->sb + 5, glob->sb, 36 * sizeof(*glob->sb));
-
- for (x=4; x >= 0; x--)
- glob->sb[x] = -scalar_product_float(glob->sb + x + 1, glob->pr1, 36);
+ memmove(ractx->sp_hist + 70, ractx->sp_hist + 75, 36*sizeof(*block));
- /* convert log and do rms */
- sum = 32. - scalar_product_float(glob->pr2, glob->lhist, 10);
+ /* block 46 of G.728 spec */
+ sum = 32.;
+ for (i=0; i < 10; i++)
+ sum -= gain_block[9-i] * ractx->gain_lpc[i];
+ /* block 47 of G.728 spec */
sum = av_clipf(sum, 0, 60);
- sumsum = exp(sum * 0.1151292546497) * gain; /* pow(10.0,sum/20)*f */
+ /* block 48 of G.728 spec */
+ /* exp(sum * 0.1151292546497) == pow(10.0,sum/20) */
+ sumsum = exp(sum * 0.1151292546497) * gain * (1.0/(1<<23));
- for (x=0; x < 5; x++)
- buffer[x] = codetable[cb_coef][x] * sumsum;
+ for (i=0; i < 5; i++)
+ buffer[i] = codetable[cb_coef][i] * sumsum;
- sum = scalar_product_float(buffer, buffer, 5) / 5;
+ sum = ff_dot_productf(buffer, buffer, 5) * ((1<<24)/5.);
sum = FFMAX(sum, 1);
/* shift and store */
- memmove(glob->lhist, glob->lhist - 1, 10 * sizeof(*glob->lhist));
-
- *glob->lhist = glob->history[glob->phase] = 10 * log10(sum) - 32;
+ memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));
- for (x=1; x < 5; x++)
- for (y=x-1; y >= 0; y--)
- buffer[x] -= glob->pr1[x-y-1] * buffer[y];
+ gain_block[9] = 10 * log10(sum) - 32;
- /* output */
- for (x=0; x < 5; x++) {
- glob->output[glob->phase*5+x] = glob->sb[4-x] =
- av_clipf(glob->sb[4-x] + buffer[x], -4095, 4095);
- }
+ ff_celp_lp_synthesis_filterf(block, ractx->sp_lpc, buffer, 5, 36);
}
-/* column multiply */
-static void colmult(float *tgt, const float *m1, const float *m2, int n)
-{
- while (n--)
- *(tgt++) = (*(m1++)) * (*(m2++));
-}
-
-static int pred(const float *in, float *tgt, int n)
+/**
+ * Hybrid window filtering, see blocks 36 and 49 of the G.728 specification.
+ *
+ * @param order filter order
+ * @param n input length
+ * @param non_rec number of non-recursive samples
+ * @param out filter output
+ * @param hist pointer to the input history of the filter
+ * @param out pointer to the non-recursive part of the output
+ * @param out2 pointer to the recursive part of the output
+ * @param window pointer to the windowing function table
+ */
+static void do_hybrid_window(RA288Context *ractx,
+ int order, int n, int non_rec, float *out,
+ float *hist, float *out2, const float *window)
{
- int x, y;
- double f0, f1, f2;
-
- if (in[n] == 0)
- return 0;
-
- if ((f0 = *in) <= 0)
- return 0;
-
- in--; // To avoid a -1 subtraction in the inner loop
-
- for (x=1; x <= n; x++) {
- f1 = in[x+1];
-
- for (y=0; y < x - 1; y++)
- f1 += in[x-y]*tgt[y];
-
- tgt[x-1] = f2 = -f1/f0;
- for (y=0; y < x >> 1; y++) {
- float temp = tgt[y] + tgt[x-y-2]*f2;
- tgt[x-y-2] += tgt[y]*f2;
- tgt[y] = temp;
- }
- if ((f0 += f1*f2) < 0)
- return 0;
- }
+ int i;
+ float buffer1[MAX_BACKWARD_FILTER_ORDER + 1];
+ float buffer2[MAX_BACKWARD_FILTER_ORDER + 1];
+ LOCAL_ALIGNED(32, float, work, [FFALIGN(MAX_BACKWARD_FILTER_ORDER +
+ MAX_BACKWARD_FILTER_LEN +
+ MAX_BACKWARD_FILTER_NONREC, 16)]);
- return 1;
-}
+ ractx->fdsp.vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
-/* product sum (lsf) */
-static void prodsum(float *tgt, const float *src, int len, int n)
-{
- for (; n >= 0; n--)
- tgt[n] = scalar_product_float(src, src - n, len);
+ convolve(buffer1, work + order , n , order);
+ convolve(buffer2, work + order + n, non_rec, order);
-}
-
-static void co(int n, int i, int j, const float *in, float *out, float *st1,
- float *st2, const float *table)
-{
- int a, b, c;
- unsigned int x;
- const float *fp;
- float buffer1[37];
- float buffer2[37];
- float work[111];
-
- /* rotate and multiply */
- c = (b = (a = n + i) + j) - i;
- fp = st1 + i;
- for (x=0; x < b; x++) {
- if (x == c)
- fp=in;
- work[x] = *(table++) * (*(st1++) = *(fp++));
+ for (i=0; i <= order; i++) {
+ out2[i] = out2[i] * 0.5625 + buffer1[i];
+ out [i] = out2[i] + buffer2[i];
}
- prodsum(buffer1, work + n, i, n);
- prodsum(buffer2, work + a, j, n);
-
- for (x=0;x<=n;x++) {
- *st2 = *st2 * (0.5625) + buffer1[x];
- out[x] = *(st2++) + buffer2[x];
- }
- *out *= 1.00390625; /* to prevent clipping */
+ /* Multiply by the white noise correcting factor (WNCF). */
+ *out *= 257./256.;
}
-static void update(Real288_internal *glob)
+/**
+ * Backward synthesis filter, find the LPC coefficients from past speech data.
+ */
+static void backward_filter(RA288Context *ractx,
+ float *hist, float *rec, const float *window,
+ float *lpc, const float *tab,
+ int order, int n, int non_rec, int move_size)
{
- float buffer1[40], temp1[37];
- float buffer2[8], temp2[11];
+ float temp[MAX_BACKWARD_FILTER_ORDER+1];
- memcpy(buffer1 , glob->output + 20, 20*sizeof(*buffer1));
- memcpy(buffer1 + 20, glob->output , 20*sizeof(*buffer1));
+ do_hybrid_window(ractx, order, n, non_rec, temp, hist, rec, window);
- co(36, 40, 35, buffer1, temp1, glob->st1a, glob->st1b, table1);
+ if (!compute_lpc_coefs(temp, order, lpc, 0, 1, 1))
+ ractx->fdsp.vector_fmul(lpc, lpc, tab, FFALIGN(order, 16));
- if (pred(temp1, glob->st1, 36))
- colmult(glob->pr1, glob->st1, table1a, 36);
-
- memcpy(buffer2 , glob->history + 4, 4*sizeof(*buffer2));
- memcpy(buffer2 + 4, glob->history , 4*sizeof(*buffer2));
-
- co(10, 8, 20, buffer2, temp2, glob->st2a, glob->st2b, table2);
-
- if (pred(temp2, glob->st2, 10))
- colmult(glob->pr2, glob->st2, table2a, 10);
+ memmove(hist, hist + n, move_size*sizeof(*hist));
}
-/* Decode a block (celp) */
static int ra288_decode_frame(AVCodecContext * avctx, void *data,
- int *data_size, const uint8_t * buf,
- int buf_size)
+ int *got_frame_ptr, AVPacket *avpkt)
{
- int16_t *out = data;
- int x, y;
- Real288_internal *glob = avctx->priv_data;
+ const uint8_t *buf = avpkt->data;
+ int buf_size = avpkt->size;
+ float *out;
+ int i, ret;
+ RA288Context *ractx = avctx->priv_data;
GetBitContext gb;
if (buf_size < avctx->block_align) {
av_log(avctx, AV_LOG_ERROR,
"Error! Input buffer is too small [%d<%d]\n",
buf_size, avctx->block_align);
- return 0;
+ return AVERROR_INVALIDDATA;
+ }
+
+ /* get output buffer */
+ ractx->frame.nb_samples = RA288_BLOCK_SIZE * RA288_BLOCKS_PER_FRAME;
+ if ((ret = avctx->get_buffer(avctx, &ractx->frame)) < 0) {
+ av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
+ return ret;
}
+ out = (float *)ractx->frame.data[0];
init_get_bits(&gb, buf, avctx->block_align * 8);
- for (x=0; x < 32; x++) {
+ for (i=0; i < RA288_BLOCKS_PER_FRAME; i++) {
float gain = amptable[get_bits(&gb, 3)];
- int cb_coef = get_bits(&gb, 6 + (x&1));
- glob->phase = x & 7;
- decode(glob, gain, cb_coef);
+ int cb_coef = get_bits(&gb, 6 + (i&1));
- for (y=0; y < 5; y++)
- *(out++) = 8 * glob->output[glob->phase*5 + y];
+ decode(ractx, gain, cb_coef);
- if (glob->phase == 3)
- update(glob);
+ memcpy(out, &ractx->sp_hist[70 + 36], RA288_BLOCK_SIZE * sizeof(*out));
+ out += RA288_BLOCK_SIZE;
+
+ if ((i & 7) == 3) {
+ backward_filter(ractx, ractx->sp_hist, ractx->sp_rec, syn_window,
+ ractx->sp_lpc, syn_bw_tab, 36, 40, 35, 70);
+
+ backward_filter(ractx, ractx->gain_hist, ractx->gain_rec, gain_window,
+ ractx->gain_lpc, gain_bw_tab, 10, 8, 20, 28);
+ }
}
- *data_size = (char *)out - (char *)data;
+ *got_frame_ptr = 1;
+ *(AVFrame *)data = ractx->frame;
+
return avctx->block_align;
}
-AVCodec ra_288_decoder =
-{
- "real_288",
- CODEC_TYPE_AUDIO,
- CODEC_ID_RA_288,
- sizeof(Real288_internal),
- NULL,
- NULL,
- NULL,
- ra288_decode_frame,
- .long_name = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),
+AVCodec ff_ra_288_decoder = {
+ .name = "real_288",
+ .type = AVMEDIA_TYPE_AUDIO,
+ .id = CODEC_ID_RA_288,
+ .priv_data_size = sizeof(RA288Context),
+ .init = ra288_decode_init,
+ .decode = ra288_decode_frame,
+ .capabilities = CODEC_CAP_DR1,
+ .long_name = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),
};