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
23 #define ALT_BITSTREAM_READER_LE
24 #include "bitstream.h"
28 float sp_lpc[36]; ///< LPC coefficients for speech data (spec: A)
29 float gain_lpc[10]; ///< LPC coefficients for gain (spec: GB)
31 float sp_hist[111]; ///< Speech data history (spec: SB)
33 /** Speech part of the gain autocorrelation (spec: REXP) */
36 float gain_hist[38]; ///< Log-gain history (spec: SBLG)
38 /** Recursive part of the gain autocorrelation (spec: REXPLG) */
41 float sp_block[41]; ///< Speech data of four blocks (spec: STTMP)
42 float gain_block[10]; ///< Gain data of four blocks (spec: GSTATE)
45 static av_cold int ra288_decode_init(AVCodecContext *avctx)
47 avctx->sample_fmt = SAMPLE_FMT_S16;
51 static inline float scalar_product_float(const float * v1, const float * v2,
62 static void colmult(float *tgt, const float *m1, const float *m2, int n)
65 *tgt++ = *m1++ * *m2++;
68 static void decode(RA288Context *ractx, float gain, int cb_coef)
73 float *block = ractx->sp_block + 36; // Current block
75 memmove(ractx->sp_block, ractx->sp_block + 5, 36*sizeof(*ractx->sp_block));
77 for (i=0; i < 5; i++) {
79 for (j=0; j < 36; j++)
80 block[i] -= block[i-1-j]*ractx->sp_lpc[j];
83 /* block 46 of G.728 spec */
85 for (i=0; i < 10; i++)
86 sum -= ractx->gain_block[9-i] * ractx->gain_lpc[i];
88 /* block 47 of G.728 spec */
89 sum = av_clipf(sum, 0, 60);
91 /* block 48 of G.728 spec */
92 sumsum = exp(sum * 0.1151292546497) * gain; /* pow(10.0,sum/20)*gain */
95 buffer[i] = codetable[cb_coef][i] * sumsum;
97 sum = scalar_product_float(buffer, buffer, 5) / 5;
101 /* shift and store */
102 memmove(ractx->gain_block, ractx->gain_block + 1,
103 9 * sizeof(*ractx->gain_block));
105 ractx->gain_block[9] = 10 * log10(sum) - 32;
107 for (i=1; i < 5; i++)
108 for (j=i-1; j >= 0; j--)
109 buffer[i] -= ractx->sp_lpc[i-j-1] * buffer[j];
112 for (i=0; i < 5; i++)
113 block[i] = av_clipf(block[i] + buffer[i], -4095, 4095);
117 * Converts autocorrelation coefficients to LPC coefficients using the
118 * Levinson-Durbin algorithm. See blocks 37 and 50 of the G.728 specification.
120 * @return 0 if success, -1 if fail
122 static int eval_lpc_coeffs(const float *in, float *tgt, int n)
133 in--; // To avoid a -1 subtraction in the inner loop
135 for (i=1; i <= n; i++) {
138 for (j=0; j < i - 1; j++)
139 f1 += in[i-j]*tgt[j];
141 tgt[i-1] = f2 = -f1/f0;
142 for (j=0; j < i >> 1; j++) {
143 float temp = tgt[j] + tgt[i-j-2]*f2;
144 tgt[i-j-2] += tgt[j]*f2;
147 if ((f0 += f1*f2) < 0)
154 static void convolve(float *tgt, const float *src, int len, int n)
157 tgt[n] = scalar_product_float(src, src - n, len);
162 * Hybrid window filtering. See blocks 36 and 49 of the G.728 specification.
164 * @param order the order of the filter
165 * @param n the length of the input
166 * @param non_rec the number of non-recursive samples
167 * @param out the filter output
168 * @param in pointer to the input of the filter
169 * @param hist pointer to the input history of the filter. It is updated by
171 * @param out pointer to the non-recursive part of the output
172 * @param out2 pointer to the recursive part of the output
173 * @param window pointer to the windowing function table
175 static void do_hybrid_window(int order, int n, int non_rec, const float *in,
176 float *out, float *hist, float *out2,
180 float buffer1[order + 1];
181 float buffer2[order + 1];
182 float work[order + n + non_rec];
185 memmove(hist , hist + n, (order + non_rec)*sizeof(*hist));
186 memcpy (hist + order + non_rec, in , n *sizeof(*hist));
188 colmult(work, window, hist, order + n + non_rec);
190 convolve(buffer1, work + order , n , order);
191 convolve(buffer2, work + order + n, non_rec, order);
193 for (i=0; i <= order; i++) {
194 out2[i] = out2[i] * 0.5625 + buffer1[i];
195 out [i] = out2[i] + buffer2[i];
198 /* Multiply by the white noise correcting factor (WNCF) */
203 * Backward synthesis filter. Find the LPC coefficients from past speech data.
205 static void backward_filter(RA288Context *ractx)
207 float temp1[37]; // RTMP in the spec
208 float temp2[11]; // GPTPMP in the spec
210 do_hybrid_window(36, 40, 35, ractx->sp_block+1, temp1, ractx->sp_hist,
211 ractx->sp_rec, syn_window);
213 if (!eval_lpc_coeffs(temp1, ractx->sp_lpc, 36))
214 colmult(ractx->sp_lpc, ractx->sp_lpc, syn_bw_tab, 36);
216 do_hybrid_window(10, 8, 20, ractx->gain_block+2, temp2, ractx->gain_hist,
217 ractx->gain_rec, gain_window);
219 if (!eval_lpc_coeffs(temp2, ractx->gain_lpc, 10))
220 colmult(ractx->gain_lpc, ractx->gain_lpc, gain_bw_tab, 10);
223 static int ra288_decode_frame(AVCodecContext * avctx, void *data,
224 int *data_size, const uint8_t * buf,
229 RA288Context *ractx = avctx->priv_data;
232 if (buf_size < avctx->block_align) {
233 av_log(avctx, AV_LOG_ERROR,
234 "Error! Input buffer is too small [%d<%d]\n",
235 buf_size, avctx->block_align);
239 init_get_bits(&gb, buf, avctx->block_align * 8);
241 for (i=0; i < 32; i++) {
242 float gain = amptable[get_bits(&gb, 3)];
243 int cb_coef = get_bits(&gb, 6 + (i&1));
245 decode(ractx, gain, cb_coef);
247 for (j=0; j < 5; j++)
248 *(out++) = 8 * ractx->sp_block[36 + j];
251 backward_filter(ractx);
254 *data_size = (char *)out - (char *)data;
255 return avctx->block_align;
258 AVCodec ra_288_decoder =
263 sizeof(RA288Context),
268 .long_name = NULL_IF_CONFIG_SMALL("RealAudio 2.0 (28.8K)"),