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
25 #include "libavutil/avassert.h"
31 static int encode_init(AVCodecContext * avctx){
32 WMACodecContext *s = avctx->priv_data;
33 int i, flags1, flags2;
38 if(avctx->channels > MAX_CHANNELS) {
39 av_log(avctx, AV_LOG_ERROR, "too many channels: got %i, need %i or fewer",
40 avctx->channels, MAX_CHANNELS);
41 return AVERROR(EINVAL);
44 if (avctx->sample_rate > 48000) {
45 av_log(avctx, AV_LOG_ERROR, "sample rate is too high: %d > 48kHz",
47 return AVERROR(EINVAL);
50 if(avctx->bit_rate < 24*1000) {
51 av_log(avctx, AV_LOG_ERROR, "bitrate too low: got %i, need 24000 or higher\n",
53 return AVERROR(EINVAL);
56 /* extract flag infos */
59 if (avctx->codec->id == CODEC_ID_WMAV1) {
60 extradata= av_malloc(4);
61 avctx->extradata_size= 4;
62 AV_WL16(extradata, flags1);
63 AV_WL16(extradata+2, flags2);
64 } else if (avctx->codec->id == CODEC_ID_WMAV2) {
65 extradata= av_mallocz(10);
66 avctx->extradata_size= 10;
67 AV_WL32(extradata, flags1);
68 AV_WL16(extradata+4, flags2);
71 avctx->extradata= extradata;
72 s->use_exp_vlc = flags2 & 0x0001;
73 s->use_bit_reservoir = flags2 & 0x0002;
74 s->use_variable_block_len = flags2 & 0x0004;
75 if (avctx->channels == 2)
78 ff_wma_init(avctx, flags2);
81 for(i = 0; i < s->nb_block_sizes; i++)
82 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 0, 1.0);
84 s->block_align = avctx->bit_rate * (int64_t)s->frame_len /
85 (avctx->sample_rate * 8);
86 s->block_align = FFMIN(s->block_align, MAX_CODED_SUPERFRAME_SIZE);
87 avctx->block_align = s->block_align;
88 avctx->bit_rate = avctx->block_align * 8LL * avctx->sample_rate /
90 //av_log(NULL, AV_LOG_ERROR, "%d %d %d %d\n", s->block_align, avctx->bit_rate, s->frame_len, avctx->sample_rate);
91 avctx->frame_size = avctx->delay = s->frame_len;
93 #if FF_API_OLD_ENCODE_AUDIO
94 avctx->coded_frame = &s->frame;
95 avcodec_get_frame_defaults(avctx->coded_frame);
102 static void apply_window_and_mdct(AVCodecContext * avctx, const signed short * audio, int len) {
103 WMACodecContext *s = avctx->priv_data;
104 int window_index= s->frame_len_bits - s->block_len_bits;
105 FFTContext *mdct = &s->mdct_ctx[window_index];
107 const float * win = s->windows[window_index];
108 int window_len = 1 << s->block_len_bits;
109 float n = window_len/2;
111 for (channel = 0; channel < avctx->channels; channel++) {
112 memcpy(s->output, s->frame_out[channel], sizeof(float)*window_len);
114 for (i = 0; i < len; i++, j += avctx->channels){
115 s->output[i+window_len] = audio[j] / n * win[window_len - i - 1];
116 s->frame_out[channel][i] = audio[j] / n * win[i];
118 mdct->mdct_calc(mdct, s->coefs[channel], s->output);
122 //FIXME use for decoding too
123 static void init_exp(WMACodecContext *s, int ch, const int *exp_param){
126 float v, *q, max_scale, *q_end;
128 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
129 q = s->exponents[ch];
130 q_end = q + s->block_len;
133 /* XXX: use a table */
134 v = pow(10, *exp_param++ * (1.0 / 16.0));
135 max_scale= FFMAX(max_scale, v);
141 s->max_exponent[ch] = max_scale;
144 static void encode_exp_vlc(WMACodecContext *s, int ch, const int *exp_param){
149 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits];
150 q = s->exponents[ch];
151 q_end = q + s->block_len;
152 if (s->version == 1) {
153 last_exp= *exp_param++;
154 assert(last_exp-10 >= 0 && last_exp-10 < 32);
155 put_bits(&s->pb, 5, last_exp - 10);
160 int exp = *exp_param++;
161 int code = exp - last_exp + 60;
162 assert(code >= 0 && code < 120);
163 put_bits(&s->pb, ff_aac_scalefactor_bits[code], ff_aac_scalefactor_code[code]);
164 /* XXX: use a table */
170 static int encode_block(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], int total_gain){
171 int v, bsize, ch, coef_nb_bits, parse_exponents;
173 int nb_coefs[MAX_CHANNELS];
174 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};
176 //FIXME remove duplication relative to decoder
177 if (s->use_variable_block_len) {
178 assert(0); //FIXME not implemented
180 /* fixed block len */
181 s->next_block_len_bits = s->frame_len_bits;
182 s->prev_block_len_bits = s->frame_len_bits;
183 s->block_len_bits = s->frame_len_bits;
186 s->block_len = 1 << s->block_len_bits;
187 // assert((s->block_pos + s->block_len) <= s->frame_len);
188 bsize = s->frame_len_bits - s->block_len_bits;
191 v = s->coefs_end[bsize] - s->coefs_start;
192 for(ch = 0; ch < s->nb_channels; ch++)
195 int n4 = s->block_len / 2;
196 mdct_norm = 1.0 / (float)n4;
197 if (s->version == 1) {
198 mdct_norm *= sqrt(n4);
202 if (s->nb_channels == 2) {
203 put_bits(&s->pb, 1, !!s->ms_stereo);
206 for(ch = 0; ch < s->nb_channels; ch++) {
207 s->channel_coded[ch] = 1; //FIXME only set channel_coded when needed, instead of always
208 if (s->channel_coded[ch]) {
209 init_exp(s, ch, fixed_exp);
213 for(ch = 0; ch < s->nb_channels; ch++) {
214 if (s->channel_coded[ch]) {
216 float *coefs, *exponents, mult;
219 coefs1 = s->coefs1[ch];
220 exponents = s->exponents[ch];
221 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch];
223 coefs = src_coefs[ch];
224 if (s->use_noise_coding && 0) {
225 assert(0); //FIXME not implemented
227 coefs += s->coefs_start;
229 for(i = 0;i < n; i++){
230 double t= *coefs++ / (exponents[i] * mult);
231 if(t<-32768 || t>32767)
234 coefs1[i] = lrint(t);
241 for(ch = 0; ch < s->nb_channels; ch++) {
242 int a = s->channel_coded[ch];
243 put_bits(&s->pb, 1, a);
250 for(v= total_gain-1; v>=127; v-= 127)
251 put_bits(&s->pb, 7, 127);
252 put_bits(&s->pb, 7, v);
254 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain);
256 if (s->use_noise_coding) {
257 for(ch = 0; ch < s->nb_channels; ch++) {
258 if (s->channel_coded[ch]) {
260 n = s->exponent_high_sizes[bsize];
262 put_bits(&s->pb, 1, s->high_band_coded[ch][i]= 0);
264 nb_coefs[ch] -= s->exponent_high_bands[bsize][i];
271 if (s->block_len_bits != s->frame_len_bits) {
272 put_bits(&s->pb, 1, parse_exponents);
275 if (parse_exponents) {
276 for(ch = 0; ch < s->nb_channels; ch++) {
277 if (s->channel_coded[ch]) {
278 if (s->use_exp_vlc) {
279 encode_exp_vlc(s, ch, fixed_exp);
281 assert(0); //FIXME not implemented
282 // encode_exp_lsp(s, ch);
287 assert(0); //FIXME not implemented
290 for(ch = 0; ch < s->nb_channels; ch++) {
291 if (s->channel_coded[ch]) {
294 tindex = (ch == 1 && s->ms_stereo);
295 ptr = &s->coefs1[ch][0];
296 eptr = ptr + nb_coefs[ch];
299 for(;ptr < eptr; ptr++){
302 int abs_level= FFABS(level);
304 if(abs_level <= s->coef_vlcs[tindex]->max_level){
305 if(run < s->coef_vlcs[tindex]->levels[abs_level-1])
306 code= run + s->int_table[tindex][abs_level-1];
309 assert(code < s->coef_vlcs[tindex]->n);
310 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[code], s->coef_vlcs[tindex]->huffcodes[code]);
313 if(1<<coef_nb_bits <= abs_level)
316 put_bits(&s->pb, coef_nb_bits, abs_level);
317 put_bits(&s->pb, s->frame_len_bits, run);
319 put_bits(&s->pb, 1, level < 0); //FIXME the sign is fliped somewhere
326 put_bits(&s->pb, s->coef_vlcs[tindex]->huffbits[1], s->coef_vlcs[tindex]->huffcodes[1]);
328 if (s->version == 1 && s->nb_channels >= 2) {
329 avpriv_align_put_bits(&s->pb);
335 static int encode_frame(WMACodecContext *s, float (*src_coefs)[BLOCK_MAX_SIZE], uint8_t *buf, int buf_size, int total_gain){
336 init_put_bits(&s->pb, buf, buf_size);
338 if (s->use_bit_reservoir) {
339 assert(0);//FIXME not implemented
341 if(encode_block(s, src_coefs, total_gain) < 0)
345 avpriv_align_put_bits(&s->pb);
347 return put_bits_count(&s->pb)/8 - s->block_align;
350 static int encode_superframe(AVCodecContext *avctx, AVPacket *avpkt,
351 const AVFrame *frame, int *got_packet_ptr)
353 WMACodecContext *s = avctx->priv_data;
354 const int16_t *samples = (const int16_t *)frame->data[0];
355 int i, total_gain, ret;
357 s->block_len_bits= s->frame_len_bits; //required by non variable block len
358 s->block_len = 1 << s->block_len_bits;
360 apply_window_and_mdct(avctx, samples, frame->nb_samples);
366 for(i = 0; i < s->block_len; i++) {
367 a = s->coefs[0][i]*0.5;
368 b = s->coefs[1][i]*0.5;
369 s->coefs[0][i] = a + b;
370 s->coefs[1][i] = a - b;
374 if ((ret = ff_alloc_packet2(avctx, avpkt, 2 * MAX_CODED_SUPERFRAME_SIZE)))
380 int error = encode_frame(s, s->coefs, avpkt->data, avpkt->size,
387 best = encode_frame(s, s->coefs, avpkt->data, avpkt->size, total_gain);
389 int scoreL = encode_frame(s, s->coefs, avpkt->data, avpkt->size, total_gain - i);
390 int scoreR = encode_frame(s, s->coefs, avpkt->data, avpkt->size, total_gain + i);
391 av_log(NULL, AV_LOG_ERROR, "%d %d %d (%d)\n", scoreL, best, scoreR, total_gain);
392 if(scoreL < FFMIN(best, scoreR)){
395 }else if(scoreR < best){
402 encode_frame(s, s->coefs, avpkt->data, avpkt->size, total_gain);
403 av_assert0((put_bits_count(&s->pb) & 7) == 0);
404 i= s->block_align - (put_bits_count(&s->pb)+7)/8;
407 put_bits(&s->pb, 8, 'N');
409 flush_put_bits(&s->pb);
410 av_assert0(put_bits_ptr(&s->pb) - s->pb.buf == s->block_align);
412 if (frame->pts != AV_NOPTS_VALUE)
413 avpkt->pts = frame->pts - ff_samples_to_time_base(avctx, avctx->delay);
415 avpkt->size = s->block_align;
420 AVCodec ff_wmav1_encoder = {
422 .type = AVMEDIA_TYPE_AUDIO,
423 .id = CODEC_ID_WMAV1,
424 .priv_data_size = sizeof(WMACodecContext),
426 .encode2 = 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 1"),
432 AVCodec ff_wmav2_encoder = {
434 .type = AVMEDIA_TYPE_AUDIO,
435 .id = CODEC_ID_WMAV2,
436 .priv_data_size = sizeof(WMACodecContext),
438 .encode2 = encode_superframe,
440 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
441 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"),