2 * WMA compatible encoder
3 * Copyright (c) 2007 Michael Niedermayer
5 * This file is part of Libav.
7 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
29 static int encode_init(AVCodecContext * avctx){
30 WMACodecContext *s = avctx->priv_data;
31 int i, flags1, flags2;
36 if(avctx->channels > MAX_CHANNELS) {
37 av_log(avctx, AV_LOG_ERROR, "too many channels: got %i, need %i or fewer",
38 avctx->channels, MAX_CHANNELS);
39 return AVERROR(EINVAL);
42 if (avctx->sample_rate > 48000) {
43 av_log(avctx, AV_LOG_ERROR, "sample rate is too high: %d > 48kHz",
45 return AVERROR(EINVAL);
48 if(avctx->bit_rate < 24*1000) {
49 av_log(avctx, AV_LOG_ERROR, "bitrate too low: got %i, need 24000 or higher\n",
51 return AVERROR(EINVAL);
54 /* extract flag infos */
57 if (avctx->codec->id == CODEC_ID_WMAV1) {
58 extradata= av_malloc(4);
59 avctx->extradata_size= 4;
60 AV_WL16(extradata, flags1);
61 AV_WL16(extradata+2, flags2);
62 } else if (avctx->codec->id == CODEC_ID_WMAV2) {
63 extradata= av_mallocz(10);
64 avctx->extradata_size= 10;
65 AV_WL32(extradata, flags1);
66 AV_WL16(extradata+4, flags2);
69 avctx->extradata= extradata;
70 s->use_exp_vlc = flags2 & 0x0001;
71 s->use_bit_reservoir = flags2 & 0x0002;
72 s->use_variable_block_len = flags2 & 0x0004;
73 if (avctx->channels == 2)
76 ff_wma_init(avctx, flags2);
79 for(i = 0; i < s->nb_block_sizes; i++)
80 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 0, 1.0);
82 s->block_align = avctx->bit_rate * (int64_t)s->frame_len /
83 (avctx->sample_rate * 8);
84 s->block_align = FFMIN(s->block_align, MAX_CODED_SUPERFRAME_SIZE);
85 avctx->block_align = s->block_align;
86 avctx->bit_rate = avctx->block_align * 8LL * avctx->sample_rate /
88 //av_log(NULL, AV_LOG_ERROR, "%d %d %d %d\n", s->block_align, avctx->bit_rate, s->frame_len, avctx->sample_rate);
89 avctx->frame_size= s->frame_len;
95 static void apply_window_and_mdct(AVCodecContext * avctx, const signed short * audio, int len) {
96 WMACodecContext *s = avctx->priv_data;
97 int window_index= s->frame_len_bits - s->block_len_bits;
98 FFTContext *mdct = &s->mdct_ctx[window_index];
100 const float * win = s->windows[window_index];
101 int window_len = 1 << s->block_len_bits;
102 float n = window_len/2;
104 for (channel = 0; channel < avctx->channels; channel++) {
105 memcpy(s->output, s->frame_out[channel], sizeof(float)*window_len);
107 for (i = 0; i < len; i++, j += avctx->channels){
108 s->output[i+window_len] = audio[j] / n * win[window_len - i - 1];
109 s->frame_out[channel][i] = audio[j] / n * win[i];
111 mdct->mdct_calc(mdct, s->coefs[channel], s->output);
115 //FIXME use for decoding too
116 static void init_exp(WMACodecContext *s, int ch, const int *exp_param){
119 float v, *q, max_scale, *q_end;
121 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
122 q = s->exponents[ch];
123 q_end = q + s->block_len;
126 /* XXX: use a table */
127 v = pow(10, *exp_param++ * (1.0 / 16.0));
128 max_scale= FFMAX(max_scale, v);
134 s->max_exponent[ch] = max_scale;
137 static void encode_exp_vlc(WMACodecContext *s, int ch, const int *exp_param){
142 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
143 q = s->exponents[ch];
144 q_end = q + s->block_len;
145 if (s->version == 1) {
146 last_exp= *exp_param++;
147 assert(last_exp-10 >= 0 && last_exp-10 < 32);
148 put_bits(&s->pb, 5, last_exp - 10);
153 int exp = *exp_param++;
154 int code = exp - last_exp + 60;
155 assert(code >= 0 && code < 120);
156 put_bits(&s->pb, ff_aac_scalefactor_bits[code], ff_aac_scalefactor_code[code]);
157 /* XXX: use a table */
163 static int encode_block(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], int total_gain){
164 int v, bsize, ch, coef_nb_bits, parse_exponents;
166 int nb_coefs[MAX_CHANNELS];
167 static const int fixed_exp[25]={20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20,20};
169 //FIXME remove duplication relative to decoder
170 if (s->use_variable_block_len) {
171 assert(0); //FIXME not implemented
173 /* fixed block len */
174 s->next_block_len_bits = s->frame_len_bits;
175 s->prev_block_len_bits = s->frame_len_bits;
176 s->block_len_bits = s->frame_len_bits;
179 s->block_len = 1 << s->block_len_bits;
180 // assert((s->block_pos + s->block_len) <= s->frame_len);
181 bsize = s->frame_len_bits - s->block_len_bits;
184 v = s->coefs_end[bsize] - s->coefs_start;
185 for(ch = 0; ch < s->nb_channels; ch++)
188 int n4 = s->block_len / 2;
189 mdct_norm = 1.0 / (float)n4;
190 if (s->version == 1) {
191 mdct_norm *= sqrt(n4);
195 if (s->nb_channels == 2) {
196 put_bits(&s->pb, 1, !!s->ms_stereo);
199 for(ch = 0; ch < s->nb_channels; ch++) {
200 s->channel_coded[ch] = 1; //FIXME only set channel_coded when needed, instead of always
201 if (s->channel_coded[ch]) {
202 init_exp(s, ch, fixed_exp);
206 for(ch = 0; ch < s->nb_channels; ch++) {
207 if (s->channel_coded[ch]) {
209 float *coefs, *exponents, mult;
212 coefs1 = s->coefs1[ch];
213 exponents = s->exponents[ch];
214 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
216 coefs = src_coefs[ch];
217 if (s->use_noise_coding && 0) {
218 assert(0); //FIXME not implemented
220 coefs += s->coefs_start;
222 for(i = 0;i < n; i++){
223 double t= *coefs++ / (exponents[i] * mult);
224 if(t<-32768 || t>32767)
227 coefs1[i] = lrint(t);
234 for(ch = 0; ch < s->nb_channels; ch++) {
235 int a = s->channel_coded[ch];
236 put_bits(&s->pb, 1, a);
243 for(v= total_gain-1; v>=127; v-= 127)
244 put_bits(&s->pb, 7, 127);
245 put_bits(&s->pb, 7, v);
247 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
249 if (s->use_noise_coding) {
250 for(ch = 0; ch < s->nb_channels; ch++) {
251 if (s->channel_coded[ch]) {
253 n = s->exponent_high_sizes[bsize];
255 put_bits(&s->pb, 1, s->high_band_coded[ch][i]= 0);
257 nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
264 if (s->block_len_bits != s->frame_len_bits) {
265 put_bits(&s->pb, 1, parse_exponents);
268 if (parse_exponents) {
269 for(ch = 0; ch < s->nb_channels; ch++) {
270 if (s->channel_coded[ch]) {
271 if (s->use_exp_vlc) {
272 encode_exp_vlc(s, ch, fixed_exp);
274 assert(0); //FIXME not implemented
275 // encode_exp_lsp(s, ch);
280 assert(0); //FIXME not implemented
283 for(ch = 0; ch < s->nb_channels; ch++) {
284 if (s->channel_coded[ch]) {
287 tindex = (ch == 1 && s->ms_stereo);
288 ptr = &s->coefs1[ch][0];
289 eptr = ptr + nb_coefs[ch];
292 for(;ptr < eptr; ptr++){
295 int abs_level= FFABS(level);
297 if(abs_level <= s->coef_vlcs[tindex]->max_level){
298 if(run < s->coef_vlcs[tindex]->levels[abs_level-1])
299 code= run + s->int_table[tindex][abs_level-1];
302 assert(code < s->coef_vlcs[tindex]->n);
303 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[code], s->coef_vlcs[tindex]->huffcodes[code]);
306 if(1<<coef_nb_bits <= abs_level)
309 put_bits(&s->pb, coef_nb_bits, abs_level);
310 put_bits(&s->pb, s->frame_len_bits, run);
312 put_bits(&s->pb, 1, level < 0); //FIXME the sign is fliped somewhere
319 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[1], s->coef_vlcs[tindex]->huffcodes[1]);
321 if (s->version == 1 && s->nb_channels >= 2) {
322 avpriv_align_put_bits(&s->pb);
328 static int encode_frame(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], uint8_t *buf, int buf_size, int total_gain){
329 init_put_bits(&s->pb, buf, buf_size);
331 if (s->use_bit_reservoir) {
332 assert(0);//FIXME not implemented
334 if(encode_block(s, src_coefs, total_gain) < 0)
338 avpriv_align_put_bits(&s->pb);
340 return put_bits_count(&s->pb)/8 - s->block_align;
343 static int encode_superframe(AVCodecContext *avctx,
344 unsigned char *buf, int buf_size, void *data){
345 WMACodecContext *s = avctx->priv_data;
346 const short *samples = data;
349 s->block_len_bits= s->frame_len_bits; //required by non variable block len
350 s->block_len = 1 << s->block_len_bits;
352 apply_window_and_mdct(avctx, samples, avctx->frame_size);
358 for(i = 0; i < s->block_len; i++) {
359 a = s->coefs[0][i]*0.5;
360 b = s->coefs[1][i]*0.5;
361 s->coefs[0][i] = a + b;
362 s->coefs[1][i] = a - b;
366 if (buf_size < 2 * MAX_CODED_SUPERFRAME_SIZE) {
367 av_log(avctx, AV_LOG_ERROR, "output buffer size is too small\n");
368 return AVERROR(EINVAL);
374 int error= encode_frame(s, s->coefs, buf, buf_size, total_gain-i);
380 best= encode_frame(s, s->coefs, buf, buf_size, total_gain);
382 int scoreL= encode_frame(s, s->coefs, buf, buf_size, total_gain-i);
383 int scoreR= encode_frame(s, s->coefs, buf, buf_size, total_gain+i);
384 av_log(NULL, AV_LOG_ERROR, "%d %d %d (%d)\n", scoreL, best, scoreR, total_gain);
385 if(scoreL < FFMIN(best, scoreR)){
388 }else if(scoreR < best){
395 if ((i = encode_frame(s, s->coefs, buf, buf_size, total_gain)) >= 0) {
396 av_log(avctx, AV_LOG_ERROR, "required frame size too large. please "
397 "use a higher bit rate.\n");
398 return AVERROR(EINVAL);
400 assert((put_bits_count(&s->pb) & 7) == 0);
402 put_bits(&s->pb, 8, 'N');
404 flush_put_bits(&s->pb);
405 return s->block_align;
408 AVCodec ff_wmav1_encoder = {
410 .type = AVMEDIA_TYPE_AUDIO,
411 .id = CODEC_ID_WMAV1,
412 .priv_data_size = sizeof(WMACodecContext),
414 .encode = encode_superframe,
416 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
417 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
420 AVCodec ff_wmav2_encoder = {
422 .type = AVMEDIA_TYPE_AUDIO,
423 .id = CODEC_ID_WMAV2,
424 .priv_data_size = sizeof(WMACodecContext),
426 .encode = encode_superframe,
428 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
429 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),