2 * RealAudio 2.0 (28.8K)
3 * Copyright (c) 2003 The FFmpeg project
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "libavutil/channel_layout.h"
23 #include "libavutil/float_dsp.h"
24 #include "libavutil/internal.h"
26 #define BITSTREAM_READER_LE
28 #include "celp_filters.h"
34 #define MAX_BACKWARD_FILTER_ORDER 36
35 #define MAX_BACKWARD_FILTER_LEN 40
36 #define MAX_BACKWARD_FILTER_NONREC 35
38 #define RA288_BLOCK_SIZE 5
39 #define RA288_BLOCKS_PER_FRAME 32
41 typedef struct RA288Context {
42 void (*vector_fmul)(float *dst, const float *src0, const float *src1,
44 DECLARE_ALIGNED(32, float, sp_lpc)[FFALIGN(36, 16)]; ///< LPC coefficients for speech data (spec: A)
45 DECLARE_ALIGNED(32, float, gain_lpc)[FFALIGN(10, 16)]; ///< LPC coefficients for gain (spec: GB)
47 /** speech data history (spec: SB).
48 * Its first 70 coefficients are updated only at backward filtering.
52 /// speech part of the gain autocorrelation (spec: REXP)
55 /** log-gain history (spec: SBLG).
56 * Its first 28 coefficients are updated only at backward filtering.
60 /// recursive part of the gain autocorrelation (spec: REXPLG)
64 static av_cold int ra288_decode_init(AVCodecContext *avctx)
66 RA288Context *ractx = avctx->priv_data;
67 AVFloatDSPContext *fdsp;
70 avctx->channel_layout = AV_CH_LAYOUT_MONO;
71 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
73 if (avctx->block_align != 38) {
74 av_log(avctx, AV_LOG_ERROR, "unsupported block align\n");
75 return AVERROR_PATCHWELCOME;
78 fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
80 return AVERROR(ENOMEM);
81 ractx->vector_fmul = fdsp->vector_fmul;
87 static void convolve(float *tgt, const float *src, int len, int n)
90 tgt[n] = avpriv_scalarproduct_float_c(src, src - n, len);
94 static void decode(RA288Context *ractx, float gain, int cb_coef)
99 float *block = ractx->sp_hist + 70 + 36; // current block
100 float *gain_block = ractx->gain_hist + 28;
102 memmove(ractx->sp_hist + 70, ractx->sp_hist + 75, 36*sizeof(*block));
104 /* block 46 of G.728 spec */
106 for (i=0; i < 10; i++)
107 sum -= gain_block[9-i] * ractx->gain_lpc[i];
109 /* block 47 of G.728 spec */
110 sum = av_clipf(sum, 0, 60);
112 /* block 48 of G.728 spec */
113 /* exp(sum * 0.1151292546497) == pow(10.0,sum/20) */
114 sumsum = exp(sum * 0.1151292546497) * gain * (1.0/(1<<23));
116 for (i=0; i < 5; i++)
117 buffer[i] = codetable[cb_coef][i] * sumsum;
119 sum = avpriv_scalarproduct_float_c(buffer, buffer, 5);
121 sum = FFMAX(sum, 5.0 / (1<<24));
123 /* shift and store */
124 memmove(gain_block, gain_block + 1, 9 * sizeof(*gain_block));
126 gain_block[9] = 10 * log10(sum) + (10*log10(((1<<24)/5.)) - 32);
128 ff_celp_lp_synthesis_filterf(block, ractx->sp_lpc, buffer, 5, 36);
132 * Hybrid window filtering, see blocks 36 and 49 of the G.728 specification.
134 * @param order filter order
135 * @param n input length
136 * @param non_rec number of non-recursive samples
137 * @param out filter output
138 * @param hist pointer to the input history of the filter
139 * @param out pointer to the non-recursive part of the output
140 * @param out2 pointer to the recursive part of the output
141 * @param window pointer to the windowing function table
143 static void do_hybrid_window(RA288Context *ractx,
144 int order, int n, int non_rec, float *out,
145 float *hist, float *out2, const float *window)
148 float buffer1[MAX_BACKWARD_FILTER_ORDER + 1];
149 float buffer2[MAX_BACKWARD_FILTER_ORDER + 1];
150 LOCAL_ALIGNED(32, float, work, [FFALIGN(MAX_BACKWARD_FILTER_ORDER +
151 MAX_BACKWARD_FILTER_LEN +
152 MAX_BACKWARD_FILTER_NONREC, 16)]);
154 av_assert2(order>=0);
156 ractx->vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
158 convolve(buffer1, work + order , n , order);
159 convolve(buffer2, work + order + n, non_rec, order);
161 for (i=0; i <= order; i++) {
162 out2[i] = out2[i] * 0.5625 + buffer1[i];
163 out [i] = out2[i] + buffer2[i];
166 /* Multiply by the white noise correcting factor (WNCF). */
167 *out *= 257.0 / 256.0;
171 * Backward synthesis filter, find the LPC coefficients from past speech data.
173 static void backward_filter(RA288Context *ractx,
174 float *hist, float *rec, const float *window,
175 float *lpc, const float *tab,
176 int order, int n, int non_rec, int move_size)
178 float temp[MAX_BACKWARD_FILTER_ORDER+1];
180 do_hybrid_window(ractx, order, n, non_rec, temp, hist, rec, window);
182 if (!compute_lpc_coefs(temp, order, lpc, 0, 1, 1))
183 ractx->vector_fmul(lpc, lpc, tab, FFALIGN(order, 16));
185 memmove(hist, hist + n, move_size*sizeof(*hist));
188 static int ra288_decode_frame(AVCodecContext * avctx, void *data,
189 int *got_frame_ptr, AVPacket *avpkt)
191 AVFrame *frame = data;
192 const uint8_t *buf = avpkt->data;
193 int buf_size = avpkt->size;
196 RA288Context *ractx = avctx->priv_data;
199 if (buf_size < avctx->block_align) {
200 av_log(avctx, AV_LOG_ERROR,
201 "Error! Input buffer is too small [%d<%d]\n",
202 buf_size, avctx->block_align);
203 return AVERROR_INVALIDDATA;
206 ret = init_get_bits8(&gb, buf, avctx->block_align);
210 /* get output buffer */
211 frame->nb_samples = RA288_BLOCK_SIZE * RA288_BLOCKS_PER_FRAME;
212 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
214 out = (float *)frame->data[0];
216 for (i=0; i < RA288_BLOCKS_PER_FRAME; i++) {
217 float gain = amptable[get_bits(&gb, 3)];
218 int cb_coef = get_bits(&gb, 6 + (i&1));
220 decode(ractx, gain, cb_coef);
222 memcpy(out, &ractx->sp_hist[70 + 36], RA288_BLOCK_SIZE * sizeof(*out));
223 out += RA288_BLOCK_SIZE;
226 backward_filter(ractx, ractx->sp_hist, ractx->sp_rec, syn_window,
227 ractx->sp_lpc, syn_bw_tab, 36, 40, 35, 70);
229 backward_filter(ractx, ractx->gain_hist, ractx->gain_rec, gain_window,
230 ractx->gain_lpc, gain_bw_tab, 10, 8, 20, 28);
236 return avctx->block_align;
239 AVCodec ff_ra_288_decoder = {
241 .long_name = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),
242 .type = AVMEDIA_TYPE_AUDIO,
243 .id = AV_CODEC_ID_RA_288,
244 .priv_data_size = sizeof(RA288Context),
245 .init = ra288_decode_init,
246 .decode = ra288_decode_frame,
247 .capabilities = AV_CODEC_CAP_DR1,