2 * WMA compatible encoder
3 * Copyright (c) 2007 Michael Niedermayer
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
24 #include "libavutil/avassert.h"
30 static int encode_init(AVCodecContext * avctx){
31 WMACodecContext *s = avctx->priv_data;
32 int i, flags1, flags2;
37 if(avctx->channels > MAX_CHANNELS) {
38 av_log(avctx, AV_LOG_ERROR, "too many channels: got %i, need %i or fewer",
39 avctx->channels, MAX_CHANNELS);
40 return AVERROR(EINVAL);
43 if (avctx->sample_rate > 48000) {
44 av_log(avctx, AV_LOG_ERROR, "sample rate is too high: %d > 48kHz",
46 return AVERROR(EINVAL);
49 if(avctx->bit_rate < 24*1000) {
50 av_log(avctx, AV_LOG_ERROR, "bitrate too low: got %i, need 24000 or higher\n",
52 return AVERROR(EINVAL);
55 /* extract flag infos */
58 if (avctx->codec->id == CODEC_ID_WMAV1) {
59 extradata= av_malloc(4);
60 avctx->extradata_size= 4;
61 AV_WL16(extradata, flags1);
62 AV_WL16(extradata+2, flags2);
63 } else if (avctx->codec->id == CODEC_ID_WMAV2) {
64 extradata= av_mallocz(10);
65 avctx->extradata_size= 10;
66 AV_WL32(extradata, flags1);
67 AV_WL16(extradata+4, flags2);
70 avctx->extradata= extradata;
71 s->use_exp_vlc = flags2 & 0x0001;
72 s->use_bit_reservoir = flags2 & 0x0002;
73 s->use_variable_block_len = flags2 & 0x0004;
74 if (avctx->channels == 2)
77 ff_wma_init(avctx, flags2);
80 for(i = 0; i < s->nb_block_sizes; i++)
81 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 0, 1.0);
83 s->block_align = avctx->bit_rate * (int64_t)s->frame_len /
84 (avctx->sample_rate * 8);
85 s->block_align = FFMIN(s->block_align, MAX_CODED_SUPERFRAME_SIZE);
86 avctx->block_align = s->block_align;
87 avctx->bit_rate = avctx->block_align * 8LL * avctx->sample_rate /
89 //av_log(NULL, AV_LOG_ERROR, "%d %d %d %d\n", s->block_align, avctx->bit_rate, s->frame_len, avctx->sample_rate);
90 avctx->frame_size= s->frame_len;
96 static void apply_window_and_mdct(AVCodecContext * avctx, const signed short * audio, int len) {
97 WMACodecContext *s = avctx->priv_data;
98 int window_index= s->frame_len_bits - s->block_len_bits;
99 FFTContext *mdct = &s->mdct_ctx[window_index];
101 const float * win = s->windows[window_index];
102 int window_len = 1 << s->block_len_bits;
103 float n = window_len/2;
105 for (channel = 0; channel < avctx->channels; channel++) {
106 memcpy(s->output, s->frame_out[channel], sizeof(float)*window_len);
108 for (i = 0; i < len; i++, j += avctx->channels){
109 s->output[i+window_len] = audio[j] / n * win[window_len - i - 1];
110 s->frame_out[channel][i] = audio[j] / n * win[i];
112 mdct->mdct_calc(mdct, s->coefs[channel], s->output);
116 //FIXME use for decoding too
117 static void init_exp(WMACodecContext *s, int ch, const int *exp_param){
120 float v, *q, max_scale, *q_end;
122 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
123 q = s->exponents[ch];
124 q_end = q + s->block_len;
127 /* XXX: use a table */
128 v = pow(10, *exp_param++ * (1.0 / 16.0));
129 max_scale= FFMAX(max_scale, v);
135 s->max_exponent[ch] = max_scale;
138 static void encode_exp_vlc(WMACodecContext *s, int ch, const int *exp_param){
143 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
144 q = s->exponents[ch];
145 q_end = q + s->block_len;
146 if (s->version == 1) {
147 last_exp= *exp_param++;
148 assert(last_exp-10 >= 0 && last_exp-10 < 32);
149 put_bits(&s->pb, 5, last_exp - 10);
154 int exp = *exp_param++;
155 int code = exp - last_exp + 60;
156 assert(code >= 0 && code < 120);
157 put_bits(&s->pb, ff_aac_scalefactor_bits[code], ff_aac_scalefactor_code[code]);
158 /* XXX: use a table */
164 static int encode_block(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], int total_gain){
165 int v, bsize, ch, coef_nb_bits, parse_exponents;
167 int nb_coefs[MAX_CHANNELS];
168 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};
170 //FIXME remove duplication relative to decoder
171 if (s->use_variable_block_len) {
172 assert(0); //FIXME not implemented
174 /* fixed block len */
175 s->next_block_len_bits = s->frame_len_bits;
176 s->prev_block_len_bits = s->frame_len_bits;
177 s->block_len_bits = s->frame_len_bits;
180 s->block_len = 1 << s->block_len_bits;
181 // assert((s->block_pos + s->block_len) <= s->frame_len);
182 bsize = s->frame_len_bits - s->block_len_bits;
185 v = s->coefs_end[bsize] - s->coefs_start;
186 for(ch = 0; ch < s->nb_channels; ch++)
189 int n4 = s->block_len / 2;
190 mdct_norm = 1.0 / (float)n4;
191 if (s->version == 1) {
192 mdct_norm *= sqrt(n4);
196 if (s->nb_channels == 2) {
197 put_bits(&s->pb, 1, !!s->ms_stereo);
200 for(ch = 0; ch < s->nb_channels; ch++) {
201 s->channel_coded[ch] = 1; //FIXME only set channel_coded when needed, instead of always
202 if (s->channel_coded[ch]) {
203 init_exp(s, ch, fixed_exp);
207 for(ch = 0; ch < s->nb_channels; ch++) {
208 if (s->channel_coded[ch]) {
210 float *coefs, *exponents, mult;
213 coefs1 = s->coefs1[ch];
214 exponents = s->exponents[ch];
215 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
217 coefs = src_coefs[ch];
218 if (s->use_noise_coding && 0) {
219 assert(0); //FIXME not implemented
221 coefs += s->coefs_start;
223 for(i = 0;i < n; i++){
224 double t= *coefs++ / (exponents[i] * mult);
225 if(t<-32768 || t>32767)
228 coefs1[i] = lrint(t);
235 for(ch = 0; ch < s->nb_channels; ch++) {
236 int a = s->channel_coded[ch];
237 put_bits(&s->pb, 1, a);
244 for(v= total_gain-1; v>=127; v-= 127)
245 put_bits(&s->pb, 7, 127);
246 put_bits(&s->pb, 7, v);
248 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
250 if (s->use_noise_coding) {
251 for(ch = 0; ch < s->nb_channels; ch++) {
252 if (s->channel_coded[ch]) {
254 n = s->exponent_high_sizes[bsize];
256 put_bits(&s->pb, 1, s->high_band_coded[ch][i]= 0);
258 nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
265 if (s->block_len_bits != s->frame_len_bits) {
266 put_bits(&s->pb, 1, parse_exponents);
269 if (parse_exponents) {
270 for(ch = 0; ch < s->nb_channels; ch++) {
271 if (s->channel_coded[ch]) {
272 if (s->use_exp_vlc) {
273 encode_exp_vlc(s, ch, fixed_exp);
275 assert(0); //FIXME not implemented
276 // encode_exp_lsp(s, ch);
281 assert(0); //FIXME not implemented
284 for(ch = 0; ch < s->nb_channels; ch++) {
285 if (s->channel_coded[ch]) {
288 tindex = (ch == 1 && s->ms_stereo);
289 ptr = &s->coefs1[ch][0];
290 eptr = ptr + nb_coefs[ch];
293 for(;ptr < eptr; ptr++){
296 int abs_level= FFABS(level);
298 if(abs_level <= s->coef_vlcs[tindex]->max_level){
299 if(run < s->coef_vlcs[tindex]->levels[abs_level-1])
300 code= run + s->int_table[tindex][abs_level-1];
303 assert(code < s->coef_vlcs[tindex]->n);
304 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[code], s->coef_vlcs[tindex]->huffcodes[code]);
307 if(1<<coef_nb_bits <= abs_level)
311 //Workaround minor rounding differences for the regression tests, FIXME we should find and replace the problematic float by fixpoint for reg tests
312 if(abs_level == 0x71B && (s->avctx->flags & CODEC_FLAG_BITEXACT)) abs_level=0x71A;
314 put_bits(&s->pb, coef_nb_bits, abs_level);
315 put_bits(&s->pb, s->frame_len_bits, run);
317 put_bits(&s->pb, 1, level < 0); //FIXME the sign is fliped somewhere
324 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[1], s->coef_vlcs[tindex]->huffcodes[1]);
326 if (s->version == 1 && s->nb_channels >= 2) {
327 avpriv_align_put_bits(&s->pb);
333 static int encode_frame(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], uint8_t *buf, int buf_size, int total_gain){
334 init_put_bits(&s->pb, buf, buf_size);
336 if (s->use_bit_reservoir) {
337 assert(0);//FIXME not implemented
339 if(encode_block(s, src_coefs, total_gain) < 0)
343 avpriv_align_put_bits(&s->pb);
345 return put_bits_count(&s->pb)/8 - s->block_align;
348 static int encode_superframe(AVCodecContext *avctx,
349 unsigned char *buf, int buf_size, void *data){
350 WMACodecContext *s = avctx->priv_data;
351 const short *samples = data;
354 s->block_len_bits= s->frame_len_bits; //required by non variable block len
355 s->block_len = 1 << s->block_len_bits;
357 apply_window_and_mdct(avctx, samples, avctx->frame_size);
363 for(i = 0; i < s->block_len; i++) {
364 a = s->coefs[0][i]*0.5;
365 b = s->coefs[1][i]*0.5;
366 s->coefs[0][i] = a + b;
367 s->coefs[1][i] = a - b;
371 if (buf_size < 2 * MAX_CODED_SUPERFRAME_SIZE) {
372 av_log(avctx, AV_LOG_ERROR, "output buffer size is too small\n");
373 return AVERROR(EINVAL);
379 int error= encode_frame(s, s->coefs, buf, buf_size, total_gain-i);
385 best= encode_frame(s, s->coefs, buf, buf_size, total_gain);
387 int scoreL= encode_frame(s, s->coefs, buf, buf_size, total_gain-i);
388 int scoreR= encode_frame(s, s->coefs, buf, buf_size, total_gain+i);
389 av_log(NULL, AV_LOG_ERROR, "%d %d %d (%d)\n", scoreL, best, scoreR, total_gain);
390 if(scoreL < FFMIN(best, scoreR)){
393 }else if(scoreR < best){
400 encode_frame(s, s->coefs, buf, buf_size, total_gain);
401 av_assert0((put_bits_count(&s->pb) & 7) == 0);
402 i= s->block_align - (put_bits_count(&s->pb)+7)/8;
405 put_bits(&s->pb, 8, 'N');
407 flush_put_bits(&s->pb);
408 av_assert0(put_bits_ptr(&s->pb) - s->pb.buf == s->block_align);
409 return s->block_align;
412 AVCodec ff_wmav1_encoder = {
414 .type = AVMEDIA_TYPE_AUDIO,
415 .id = CODEC_ID_WMAV1,
416 .priv_data_size = sizeof(WMACodecContext),
418 .encode = encode_superframe,
420 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
421 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"),
424 AVCodec ff_wmav2_encoder = {
426 .type = AVMEDIA_TYPE_AUDIO,
427 .id = CODEC_ID_WMAV2,
428 .priv_data_size = sizeof(WMACodecContext),
430 .encode = encode_superframe,
432 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
433 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),