• Main Page
  • Related Pages
  • Modules
  • Data Structures
  • Files
  • File List
  • Globals

libavcodec/twinvq.c

Go to the documentation of this file.
00001 /*
00002  * TwinVQ decoder
00003  * Copyright (c) 2009 Vitor Sessak
00004  *
00005  * This file is part of FFmpeg.
00006  *
00007  * FFmpeg is free software; you can redistribute it and/or
00008  * modify it under the terms of the GNU Lesser General Public
00009  * License as published by the Free Software Foundation; either
00010  * version 2.1 of the License, or (at your option) any later version.
00011  *
00012  * FFmpeg is distributed in the hope that it will be useful,
00013  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00014  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00015  * Lesser General Public License for more details.
00016  *
00017  * You should have received a copy of the GNU Lesser General Public
00018  * License along with FFmpeg; if not, write to the Free Software
00019  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00020  */
00021 
00022 #include "avcodec.h"
00023 #include "get_bits.h"
00024 #include "dsputil.h"
00025 #include "fft.h"
00026 #include "lsp.h"
00027 
00028 #include <math.h>
00029 #include <stdint.h>
00030 
00031 #include "twinvq_data.h"
00032 
00033 enum FrameType {
00034     FT_SHORT = 0,  
00035     FT_MEDIUM,     
00036     FT_LONG,       
00037     FT_PPC,        
00038 };
00039 
00043 struct FrameMode {
00044     uint8_t         sub;      
00045     const uint16_t *bark_tab;
00046 
00048     uint8_t         bark_env_size;
00049 
00050     const int16_t  *bark_cb;    
00051     uint8_t         bark_n_coef;
00052     uint8_t         bark_n_bit; 
00053 
00055 
00056     const int16_t    *cb0;
00057     const int16_t    *cb1;
00059 
00060     uint8_t         cb_len_read; 
00061 };
00062 
00067 typedef struct {
00068     struct FrameMode fmode[3]; 
00069 
00070     uint16_t     size;        
00071     uint8_t      n_lsp;       
00072     const float *lspcodebook;
00073 
00074     /* number of bits of the different LSP CB coefficients */
00075     uint8_t      lsp_bit0;
00076     uint8_t      lsp_bit1;
00077     uint8_t      lsp_bit2;
00078 
00079     uint8_t      lsp_split;      
00080     const int16_t *ppc_shape_cb; 
00081 
00083     uint8_t      ppc_period_bit;
00084 
00085     uint8_t      ppc_shape_bit;  
00086     uint8_t      ppc_shape_len;  
00087     uint8_t      pgain_bit;      
00088 
00090     uint16_t     peak_per2wid;
00091 } ModeTab;
00092 
00093 static const ModeTab mode_08_08 = {
00094     {
00095         { 8, bark_tab_s08_64,  10, tab.fcb08s  , 1, 5, tab.cb0808s0, tab.cb0808s1, 18},
00096         { 2, bark_tab_m08_256, 20, tab.fcb08m  , 2, 5, tab.cb0808m0, tab.cb0808m1, 16},
00097         { 1, bark_tab_l08_512, 30, tab.fcb08l  , 3, 6, tab.cb0808l0, tab.cb0808l1, 17}
00098     },
00099     512 , 12, tab.lsp08,   1, 5, 3, 3, tab.shape08  , 8, 28, 20, 6, 40
00100 };
00101 
00102 static const ModeTab mode_11_08 = {
00103     {
00104         { 8, bark_tab_s11_64,  10, tab.fcb11s  , 1, 5, tab.cb1108s0, tab.cb1108s1, 29},
00105         { 2, bark_tab_m11_256, 20, tab.fcb11m  , 2, 5, tab.cb1108m0, tab.cb1108m1, 24},
00106         { 1, bark_tab_l11_512, 30, tab.fcb11l  , 3, 6, tab.cb1108l0, tab.cb1108l1, 27}
00107     },
00108     512 , 16, tab.lsp11,   1, 6, 4, 3, tab.shape11  , 9, 36, 30, 7, 90
00109 };
00110 
00111 static const ModeTab mode_11_10 = {
00112     {
00113         { 8, bark_tab_s11_64,  10, tab.fcb11s  , 1, 5, tab.cb1110s0, tab.cb1110s1, 21},
00114         { 2, bark_tab_m11_256, 20, tab.fcb11m  , 2, 5, tab.cb1110m0, tab.cb1110m1, 18},
00115         { 1, bark_tab_l11_512, 30, tab.fcb11l  , 3, 6, tab.cb1110l0, tab.cb1110l1, 20}
00116     },
00117     512 , 16, tab.lsp11,   1, 6, 4, 3, tab.shape11  , 9, 36, 30, 7, 90
00118 };
00119 
00120 static const ModeTab mode_16_16 = {
00121     {
00122         { 8, bark_tab_s16_128, 10, tab.fcb16s  , 1, 5, tab.cb1616s0, tab.cb1616s1, 16},
00123         { 2, bark_tab_m16_512, 20, tab.fcb16m  , 2, 5, tab.cb1616m0, tab.cb1616m1, 15},
00124         { 1, bark_tab_l16_1024,30, tab.fcb16l  , 3, 6, tab.cb1616l0, tab.cb1616l1, 16}
00125     },
00126     1024, 16, tab.lsp16,   1, 6, 4, 3, tab.shape16  , 9, 56, 60, 7, 180
00127 };
00128 
00129 static const ModeTab mode_22_20 = {
00130     {
00131         { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2220s0, tab.cb2220s1, 18},
00132         { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2220m0, tab.cb2220m1, 17},
00133         { 1, bark_tab_l22_1024,32, tab.fcb22l_1, 4, 6, tab.cb2220l0, tab.cb2220l1, 18}
00134     },
00135     1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
00136 };
00137 
00138 static const ModeTab mode_22_24 = {
00139     {
00140         { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2224s0, tab.cb2224s1, 15},
00141         { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2224m0, tab.cb2224m1, 14},
00142         { 1, bark_tab_l22_1024,32, tab.fcb22l_1, 4, 6, tab.cb2224l0, tab.cb2224l1, 15}
00143     },
00144     1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
00145 };
00146 
00147 static const ModeTab mode_22_32 = {
00148     {
00149         { 4, bark_tab_s22_128, 10, tab.fcb22s_2, 1, 6, tab.cb2232s0, tab.cb2232s1, 11},
00150         { 2, bark_tab_m22_256, 20, tab.fcb22m_2, 2, 6, tab.cb2232m0, tab.cb2232m1, 11},
00151         { 1, bark_tab_l22_512, 32, tab.fcb22l_2, 4, 6, tab.cb2232l0, tab.cb2232l1, 12}
00152     },
00153     512 , 16, tab.lsp22_2, 1, 6, 4, 4, tab.shape22_2, 9, 56, 36, 7, 72
00154 };
00155 
00156 static const ModeTab mode_44_40 = {
00157     {
00158         {16, bark_tab_s44_128, 10, tab.fcb44s  , 1, 6, tab.cb4440s0, tab.cb4440s1, 18},
00159         { 4, bark_tab_m44_512, 20, tab.fcb44m  , 2, 6, tab.cb4440m0, tab.cb4440m1, 17},
00160         { 1, bark_tab_l44_2048,40, tab.fcb44l  , 4, 6, tab.cb4440l0, tab.cb4440l1, 17}
00161     },
00162     2048, 20, tab.lsp44,   1, 6, 4, 4, tab.shape44  , 9, 84, 54, 7, 432
00163 };
00164 
00165 static const ModeTab mode_44_48 = {
00166     {
00167         {16, bark_tab_s44_128, 10, tab.fcb44s  , 1, 6, tab.cb4448s0, tab.cb4448s1, 15},
00168         { 4, bark_tab_m44_512, 20, tab.fcb44m  , 2, 6, tab.cb4448m0, tab.cb4448m1, 14},
00169         { 1, bark_tab_l44_2048,40, tab.fcb44l  , 4, 6, tab.cb4448l0, tab.cb4448l1, 14}
00170     },
00171     2048, 20, tab.lsp44,   1, 6, 4, 4, tab.shape44  , 9, 84, 54, 7, 432
00172 };
00173 
00174 typedef struct TwinContext {
00175     AVCodecContext *avctx;
00176     DSPContext      dsp;
00177     FFTContext mdct_ctx[3];
00178 
00179     const ModeTab *mtab;
00180 
00181     // history
00182     float lsp_hist[2][20];           
00183     float bark_hist[3][2][40];       
00184 
00185     // bitstream parameters
00186     int16_t permut[4][4096];
00187     uint8_t length[4][2];            
00188     uint8_t length_change[4];
00189     uint8_t bits_main_spec[2][4][2]; 
00190     int bits_main_spec_change[4];
00191     int n_div[4];
00192 
00193     float *spectrum;
00194     float *curr_frame;               
00195     float *prev_frame;               
00196     int last_block_pos[2];
00197 
00198     float *cos_tabs[3];
00199 
00200     // scratch buffers
00201     float *tmp_buf;
00202 } TwinContext;
00203 
00204 #define PPC_SHAPE_CB_SIZE 64
00205 #define SUB_AMP_MAX       4500.0
00206 #define MULAW_MU          100.0
00207 #define GAIN_BITS         8
00208 #define AMP_MAX           13000.0
00209 #define SUB_GAIN_BITS     5
00210 #define WINDOW_TYPE_BITS  4
00211 #define PGAIN_MU          200
00212 
00214 static void memset_float(float *buf, float val, int size)
00215 {
00216     while (size--)
00217         *buf++ = val;
00218 }
00219 
00232 static float eval_lpc_spectrum(const float *lsp, float cos_val, int order)
00233 {
00234     int j;
00235     float p = 0.5f;
00236     float q = 0.5f;
00237     float two_cos_w = 2.0f*cos_val;
00238 
00239     for (j = 0; j + 1 < order; j += 2*2) {
00240         // Unroll the loop once since order is a multiple of four
00241         q *= lsp[j  ] - two_cos_w;
00242         p *= lsp[j+1] - two_cos_w;
00243 
00244         q *= lsp[j+2] - two_cos_w;
00245         p *= lsp[j+3] - two_cos_w;
00246     }
00247 
00248     p *= p * (2.0f - two_cos_w);
00249     q *= q * (2.0f + two_cos_w);
00250 
00251     return 0.5 / (p + q);
00252 }
00253 
00257 static void eval_lpcenv(TwinContext *tctx, const float *cos_vals, float *lpc)
00258 {
00259     int i;
00260     const ModeTab *mtab = tctx->mtab;
00261     int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;
00262 
00263     for (i = 0; i < size_s/2; i++) {
00264         float cos_i = tctx->cos_tabs[0][i];
00265         lpc[i]          = eval_lpc_spectrum(cos_vals,  cos_i, mtab->n_lsp);
00266         lpc[size_s-i-1] = eval_lpc_spectrum(cos_vals, -cos_i, mtab->n_lsp);
00267     }
00268 }
00269 
00270 static void interpolate(float *out, float v1, float v2, int size)
00271 {
00272     int i;
00273     float step = (v1 - v2)/(size + 1);
00274 
00275     for (i = 0; i < size; i++) {
00276         v2 += step;
00277         out[i] = v2;
00278     }
00279 }
00280 
00281 static inline float get_cos(int idx, int part, const float *cos_tab, int size)
00282 {
00283     return part ? -cos_tab[size - idx - 1] :
00284                    cos_tab[       idx    ];
00285 }
00286 
00301 static inline void eval_lpcenv_or_interp(TwinContext *tctx,
00302                                          enum FrameType ftype,
00303                                          float *out, const float *in,
00304                                          int size, int step, int part)
00305 {
00306     int i;
00307     const ModeTab *mtab = tctx->mtab;
00308     const float *cos_tab = tctx->cos_tabs[ftype];
00309 
00310     // Fill the 's'
00311     for (i = 0; i < size; i += step)
00312         out[i] =
00313             eval_lpc_spectrum(in,
00314                               get_cos(i, part, cos_tab, size),
00315                               mtab->n_lsp);
00316 
00317     // Fill the 'iiiibiiii'
00318     for (i = step; i <= size - 2*step; i += step) {
00319         if (out[i + step] + out[i - step] >  1.95*out[i] ||
00320             out[i + step]                 >=  out[i - step]) {
00321             interpolate(out + i - step + 1, out[i], out[i-step], step - 1);
00322         } else {
00323             out[i - step/2] =
00324                 eval_lpc_spectrum(in,
00325                                   get_cos(i-step/2, part, cos_tab, size),
00326                                   mtab->n_lsp);
00327             interpolate(out + i - step   + 1, out[i-step/2], out[i-step  ], step/2 - 1);
00328             interpolate(out + i - step/2 + 1, out[i       ], out[i-step/2], step/2 - 1);
00329         }
00330     }
00331 
00332     interpolate(out + size - 2*step + 1, out[size-step], out[size - 2*step], step - 1);
00333 }
00334 
00335 static void eval_lpcenv_2parts(TwinContext *tctx, enum FrameType ftype,
00336                                const float *buf, float *lpc,
00337                                int size, int step)
00338 {
00339     eval_lpcenv_or_interp(tctx, ftype, lpc         , buf, size/2,   step, 0);
00340     eval_lpcenv_or_interp(tctx, ftype, lpc + size/2, buf, size/2, 2*step, 1);
00341 
00342     interpolate(lpc+size/2-step+1, lpc[size/2], lpc[size/2-step], step);
00343 
00344     memset_float(lpc + size - 2*step + 1, lpc[size - 2*step], 2*step - 1);
00345 }
00346 
00352 static void dequant(TwinContext *tctx, GetBitContext *gb, float *out,
00353                     enum FrameType ftype,
00354                     const int16_t *cb0, const int16_t *cb1, int cb_len)
00355 {
00356     int pos = 0;
00357     int i, j;
00358 
00359     for (i = 0; i < tctx->n_div[ftype]; i++) {
00360         int tmp0, tmp1;
00361         int sign0 = 1;
00362         int sign1 = 1;
00363         const int16_t *tab0, *tab1;
00364         int length = tctx->length[ftype][i >= tctx->length_change[ftype]];
00365         int bitstream_second_part = (i >= tctx->bits_main_spec_change[ftype]);
00366 
00367         int bits = tctx->bits_main_spec[0][ftype][bitstream_second_part];
00368         if (bits == 7) {
00369             if (get_bits1(gb))
00370                 sign0 = -1;
00371             bits = 6;
00372         }
00373         tmp0 = get_bits(gb, bits);
00374 
00375         bits = tctx->bits_main_spec[1][ftype][bitstream_second_part];
00376 
00377         if (bits == 7) {
00378             if (get_bits1(gb))
00379                 sign1 = -1;
00380 
00381             bits = 6;
00382         }
00383         tmp1 = get_bits(gb, bits);
00384 
00385         tab0 = cb0 + tmp0*cb_len;
00386         tab1 = cb1 + tmp1*cb_len;
00387 
00388         for (j = 0; j < length; j++)
00389             out[tctx->permut[ftype][pos+j]] = sign0*tab0[j] + sign1*tab1[j];
00390 
00391         pos += length;
00392     }
00393 
00394 }
00395 
00396 static inline float mulawinv(float y, float clip, float mu)
00397 {
00398     y = av_clipf(y/clip, -1, 1);
00399     return clip * FFSIGN(y) * (exp(log(1+mu) * fabs(y)) - 1) / mu;
00400 }
00401 
00422 static int very_broken_op(int a, int b)
00423 {
00424     int x = a*b + 200;
00425     int size;
00426     const uint8_t *rtab;
00427 
00428     if (x%400 || b%5)
00429         return x/400;
00430 
00431     x /= 400;
00432 
00433     size = tabs[b/5].size;
00434     rtab = tabs[b/5].tab;
00435     return x - rtab[size*av_log2(2*(x - 1)/size)+(x - 1)%size];
00436 }
00437 
00443 static void add_peak(int period, int width, const float *shape,
00444                      float ppc_gain, float *speech, int len)
00445 {
00446     int i, j;
00447 
00448     const float *shape_end = shape + len;
00449     int center;
00450 
00451     // First peak centered around zero
00452     for (i = 0; i < width/2; i++)
00453         speech[i] += ppc_gain * *shape++;
00454 
00455     for (i = 1; i < ROUNDED_DIV(len,width) ; i++) {
00456         center = very_broken_op(period, i);
00457         for (j = -width/2; j < (width+1)/2; j++)
00458             speech[j+center] += ppc_gain * *shape++;
00459     }
00460 
00461     // For the last block, be careful not to go beyond the end of the buffer
00462     center = very_broken_op(period, i);
00463     for (j = -width/2; j < (width + 1)/2 && shape < shape_end; j++)
00464         speech[j+center] += ppc_gain * *shape++;
00465 }
00466 
00467 static void decode_ppc(TwinContext *tctx, int period_coef, const float *shape,
00468                        float ppc_gain, float *speech)
00469 {
00470     const ModeTab *mtab = tctx->mtab;
00471     int isampf = tctx->avctx->sample_rate/1000;
00472     int ibps = tctx->avctx->bit_rate/(1000 * tctx->avctx->channels);
00473     int min_period = ROUNDED_DIV(  40*2*mtab->size, isampf);
00474     int max_period = ROUNDED_DIV(6*40*2*mtab->size, isampf);
00475     int period_range = max_period - min_period;
00476 
00477     // This is actually the period multiplied by 400. It is just linearly coded
00478     // between its maximum and minimum value.
00479     int period = min_period +
00480         ROUNDED_DIV(period_coef*period_range, (1 << mtab->ppc_period_bit) - 1);
00481     int width;
00482 
00483     if (isampf == 22 && ibps == 32) {
00484         // For some unknown reason, NTT decided to code this case differently...
00485         width = ROUNDED_DIV((period + 800)* mtab->peak_per2wid, 400*mtab->size);
00486     } else
00487         width =             (period      )* mtab->peak_per2wid/(400*mtab->size);
00488 
00489     add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len);
00490 }
00491 
00492 static void dec_gain(TwinContext *tctx, GetBitContext *gb, enum FrameType ftype,
00493                      float *out)
00494 {
00495     const ModeTab *mtab = tctx->mtab;
00496     int i, j;
00497     int sub = mtab->fmode[ftype].sub;
00498     float step     = AMP_MAX     / ((1 <<     GAIN_BITS) - 1);
00499     float sub_step = SUB_AMP_MAX / ((1 << SUB_GAIN_BITS) - 1);
00500 
00501     if (ftype == FT_LONG) {
00502         for (i = 0; i < tctx->avctx->channels; i++)
00503             out[i] = (1./(1<<13)) *
00504                 mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
00505                          AMP_MAX, MULAW_MU);
00506     } else {
00507         for (i = 0; i < tctx->avctx->channels; i++) {
00508             float val = (1./(1<<23)) *
00509                 mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
00510                          AMP_MAX, MULAW_MU);
00511 
00512             for (j = 0; j < sub; j++) {
00513                 out[i*sub + j] =
00514                     val*mulawinv(sub_step* 0.5 +
00515                                  sub_step* get_bits(gb, SUB_GAIN_BITS),
00516                                  SUB_AMP_MAX, MULAW_MU);
00517             }
00518         }
00519     }
00520 }
00521 
00528 static void rearrange_lsp(int order, float *lsp, float min_dist)
00529 {
00530     int i;
00531     float min_dist2 = min_dist * 0.5;
00532     for (i = 1; i < order; i++)
00533         if (lsp[i] - lsp[i-1] < min_dist) {
00534             float avg = (lsp[i] + lsp[i-1]) * 0.5;
00535 
00536             lsp[i-1] = avg - min_dist2;
00537             lsp[i  ] = avg + min_dist2;
00538         }
00539 }
00540 
00541 static void decode_lsp(TwinContext *tctx, int lpc_idx1, uint8_t *lpc_idx2,
00542                        int lpc_hist_idx, float *lsp, float *hist)
00543 {
00544     const ModeTab *mtab = tctx->mtab;
00545     int i, j;
00546 
00547     const float *cb  =  mtab->lspcodebook;
00548     const float *cb2 =  cb  + (1 << mtab->lsp_bit1)*mtab->n_lsp;
00549     const float *cb3 =  cb2 + (1 << mtab->lsp_bit2)*mtab->n_lsp;
00550 
00551     const int8_t funny_rounding[4] = {
00552         -2,
00553         mtab->lsp_split == 4 ? -2 : 1,
00554         mtab->lsp_split == 4 ? -2 : 1,
00555         0
00556     };
00557 
00558     j = 0;
00559     for (i = 0; i < mtab->lsp_split; i++) {
00560         int chunk_end = ((i + 1)*mtab->n_lsp + funny_rounding[i])/mtab->lsp_split;
00561         for (; j < chunk_end; j++)
00562             lsp[j] = cb [lpc_idx1    * mtab->n_lsp + j] +
00563                      cb2[lpc_idx2[i] * mtab->n_lsp + j];
00564     }
00565 
00566     rearrange_lsp(mtab->n_lsp, lsp, 0.0001);
00567 
00568     for (i = 0; i < mtab->n_lsp; i++) {
00569         float tmp1 = 1. -          cb3[lpc_hist_idx*mtab->n_lsp + i];
00570         float tmp2 =     hist[i] * cb3[lpc_hist_idx*mtab->n_lsp + i];
00571         hist[i] = lsp[i];
00572         lsp[i]  = lsp[i] * tmp1 + tmp2;
00573     }
00574 
00575     rearrange_lsp(mtab->n_lsp, lsp, 0.0001);
00576     rearrange_lsp(mtab->n_lsp, lsp, 0.000095);
00577     ff_sort_nearly_sorted_floats(lsp, mtab->n_lsp);
00578 }
00579 
00580 static void dec_lpc_spectrum_inv(TwinContext *tctx, float *lsp,
00581                                  enum FrameType ftype, float *lpc)
00582 {
00583     int i;
00584     int size = tctx->mtab->size / tctx->mtab->fmode[ftype].sub;
00585 
00586     for (i = 0; i < tctx->mtab->n_lsp; i++)
00587         lsp[i] =  2*cos(lsp[i]);
00588 
00589     switch (ftype) {
00590     case FT_LONG:
00591         eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 8);
00592         break;
00593     case FT_MEDIUM:
00594         eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 2);
00595         break;
00596     case FT_SHORT:
00597         eval_lpcenv(tctx, lsp, lpc);
00598         break;
00599     }
00600 }
00601 
00602 static void imdct_and_window(TwinContext *tctx, enum FrameType ftype, int wtype,
00603                             float *in, float *prev, int ch)
00604 {
00605     const ModeTab *mtab = tctx->mtab;
00606     int bsize = mtab->size / mtab->fmode[ftype].sub;
00607     int size  = mtab->size;
00608     float *buf1 = tctx->tmp_buf;
00609     int j;
00610     int wsize; // Window size
00611     float *out = tctx->curr_frame + 2*ch*mtab->size;
00612     float *out2 = out;
00613     float *prev_buf;
00614     int first_wsize;
00615 
00616     static const uint8_t wtype_to_wsize[]      = {0, 0, 2, 2, 2, 1, 0, 1, 1};
00617     int types_sizes[] = {
00618         mtab->size /    mtab->fmode[FT_LONG  ].sub,
00619         mtab->size /    mtab->fmode[FT_MEDIUM].sub,
00620         mtab->size / (2*mtab->fmode[FT_SHORT ].sub),
00621     };
00622 
00623     wsize = types_sizes[wtype_to_wsize[wtype]];
00624     first_wsize = wsize;
00625     prev_buf = prev + (size - bsize)/2;
00626 
00627     for (j = 0; j < mtab->fmode[ftype].sub; j++) {
00628         int sub_wtype = ftype == FT_MEDIUM ? 8 : wtype;
00629 
00630         if (!j && wtype == 4)
00631             sub_wtype = 4;
00632         else if (j == mtab->fmode[ftype].sub-1 && wtype == 7)
00633             sub_wtype = 7;
00634 
00635         wsize = types_sizes[wtype_to_wsize[sub_wtype]];
00636 
00637         ff_imdct_half(&tctx->mdct_ctx[ftype], buf1 + bsize*j, in + bsize*j);
00638 
00639         tctx->dsp.vector_fmul_window(out2,
00640                                      prev_buf + (bsize-wsize)/2,
00641                                      buf1 + bsize*j,
00642                                      ff_sine_windows[av_log2(wsize)],
00643                                      0.0,
00644                                      wsize/2);
00645         out2 += wsize;
00646 
00647         memcpy(out2, buf1 + bsize*j + wsize/2, (bsize - wsize/2)*sizeof(float));
00648 
00649         out2 += ftype == FT_MEDIUM ? (bsize-wsize)/2 : bsize - wsize;
00650 
00651         prev_buf = buf1 + bsize*j + bsize/2;
00652     }
00653 
00654     tctx->last_block_pos[ch] = (size + first_wsize)/2;
00655 }
00656 
00657 static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype,
00658                          float *out)
00659 {
00660     const ModeTab *mtab = tctx->mtab;
00661     float *prev_buf = tctx->prev_frame + tctx->last_block_pos[0];
00662     int i, j;
00663 
00664     for (i = 0; i < tctx->avctx->channels; i++) {
00665         imdct_and_window(tctx, ftype, wtype,
00666                          tctx->spectrum + i*mtab->size,
00667                          prev_buf + 2*i*mtab->size,
00668                          i);
00669     }
00670 
00671     if (tctx->avctx->channels == 2) {
00672         for (i = 0; i < mtab->size - tctx->last_block_pos[0]; i++) {
00673             float f1 = prev_buf[               i];
00674             float f2 = prev_buf[2*mtab->size + i];
00675             out[2*i    ] = f1 + f2;
00676             out[2*i + 1] = f1 - f2;
00677         }
00678         for (j = 0; i < mtab->size; j++,i++) {
00679             float f1 = tctx->curr_frame[               j];
00680             float f2 = tctx->curr_frame[2*mtab->size + j];
00681             out[2*i    ] = f1 + f2;
00682             out[2*i + 1] = f1 - f2;
00683         }
00684     } else {
00685         memcpy(out, prev_buf,
00686                (mtab->size - tctx->last_block_pos[0]) * sizeof(*out));
00687 
00688         out +=  mtab->size - tctx->last_block_pos[0];
00689 
00690         memcpy(out, tctx->curr_frame,
00691                (tctx->last_block_pos[0]) * sizeof(*out));
00692     }
00693 
00694 }
00695 
00696 static void dec_bark_env(TwinContext *tctx, const uint8_t *in, int use_hist,
00697                          int ch, float *out, float gain, enum FrameType ftype)
00698 {
00699     const ModeTab *mtab = tctx->mtab;
00700     int i,j;
00701     float *hist = tctx->bark_hist[ftype][ch];
00702     float val = ((const float []) {0.4, 0.35, 0.28})[ftype];
00703     int bark_n_coef  = mtab->fmode[ftype].bark_n_coef;
00704     int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef;
00705     int idx = 0;
00706 
00707     for (i = 0; i < fw_cb_len; i++)
00708         for (j = 0; j < bark_n_coef; j++, idx++) {
00709             float tmp2 =
00710                 mtab->fmode[ftype].bark_cb[fw_cb_len*in[j] + i] * (1./4096);
00711             float st = use_hist ?
00712                 (1. - val) * tmp2 + val*hist[idx] + 1. : tmp2 + 1.;
00713 
00714             hist[idx] = tmp2;
00715             if (st < -1.) st = 1.;
00716 
00717             memset_float(out, st * gain, mtab->fmode[ftype].bark_tab[idx]);
00718             out += mtab->fmode[ftype].bark_tab[idx];
00719         }
00720 
00721 }
00722 
00723 static void read_and_decode_spectrum(TwinContext *tctx, GetBitContext *gb,
00724                                      float *out, enum FrameType ftype)
00725 {
00726     const ModeTab *mtab = tctx->mtab;
00727     int channels = tctx->avctx->channels;
00728     int sub = mtab->fmode[ftype].sub;
00729     int block_size = mtab->size / sub;
00730     float gain[channels*sub];
00731     float ppc_shape[mtab->ppc_shape_len * channels * 4];
00732     uint8_t bark1[channels][sub][mtab->fmode[ftype].bark_n_coef];
00733     uint8_t bark_use_hist[channels][sub];
00734 
00735     uint8_t lpc_idx1[channels];
00736     uint8_t lpc_idx2[channels][tctx->mtab->lsp_split];
00737     uint8_t lpc_hist_idx[channels];
00738 
00739     int i, j, k;
00740 
00741     dequant(tctx, gb, out, ftype,
00742             mtab->fmode[ftype].cb0, mtab->fmode[ftype].cb1,
00743             mtab->fmode[ftype].cb_len_read);
00744 
00745     for (i = 0; i < channels; i++)
00746         for (j = 0; j < sub; j++)
00747             for (k = 0; k < mtab->fmode[ftype].bark_n_coef; k++)
00748                 bark1[i][j][k] =
00749                     get_bits(gb, mtab->fmode[ftype].bark_n_bit);
00750 
00751     for (i = 0; i < channels; i++)
00752         for (j = 0; j < sub; j++)
00753             bark_use_hist[i][j] = get_bits1(gb);
00754 
00755     dec_gain(tctx, gb, ftype, gain);
00756 
00757     for (i = 0; i < channels; i++) {
00758         lpc_hist_idx[i] = get_bits(gb, tctx->mtab->lsp_bit0);
00759         lpc_idx1    [i] = get_bits(gb, tctx->mtab->lsp_bit1);
00760 
00761         for (j = 0; j < tctx->mtab->lsp_split; j++)
00762             lpc_idx2[i][j] = get_bits(gb, tctx->mtab->lsp_bit2);
00763     }
00764 
00765     if (ftype == FT_LONG) {
00766         int cb_len_p = (tctx->n_div[3] + mtab->ppc_shape_len*channels - 1)/
00767             tctx->n_div[3];
00768         dequant(tctx, gb, ppc_shape, FT_PPC, mtab->ppc_shape_cb,
00769                 mtab->ppc_shape_cb + cb_len_p*PPC_SHAPE_CB_SIZE, cb_len_p);
00770     }
00771 
00772     for (i = 0; i < channels; i++) {
00773         float *chunk = out + mtab->size * i;
00774         float lsp[tctx->mtab->n_lsp];
00775 
00776         for (j = 0; j < sub; j++) {
00777             dec_bark_env(tctx, bark1[i][j], bark_use_hist[i][j], i,
00778                          tctx->tmp_buf, gain[sub*i+j], ftype);
00779 
00780             tctx->dsp.vector_fmul(chunk + block_size*j, tctx->tmp_buf,
00781                                   block_size);
00782 
00783         }
00784 
00785         if (ftype == FT_LONG) {
00786             float pgain_step = 25000. / ((1 << mtab->pgain_bit) - 1);
00787             int p_coef = get_bits(gb, tctx->mtab->ppc_period_bit);
00788             int g_coef = get_bits(gb, tctx->mtab->pgain_bit);
00789             float v = 1./8192*
00790                 mulawinv(pgain_step*g_coef+ pgain_step/2, 25000., PGAIN_MU);
00791 
00792             decode_ppc(tctx, p_coef, ppc_shape + i*mtab->ppc_shape_len, v,
00793                        chunk);
00794         }
00795 
00796         decode_lsp(tctx, lpc_idx1[i], lpc_idx2[i], lpc_hist_idx[i], lsp,
00797                    tctx->lsp_hist[i]);
00798 
00799         dec_lpc_spectrum_inv(tctx, lsp, ftype, tctx->tmp_buf);
00800 
00801         for (j = 0; j < mtab->fmode[ftype].sub; j++) {
00802             tctx->dsp.vector_fmul(chunk, tctx->tmp_buf, block_size);
00803             chunk += block_size;
00804         }
00805     }
00806 }
00807 
00808 static int twin_decode_frame(AVCodecContext * avctx, void *data,
00809                              int *data_size, AVPacket *avpkt)
00810 {
00811     const uint8_t *buf = avpkt->data;
00812     int buf_size = avpkt->size;
00813     TwinContext *tctx = avctx->priv_data;
00814     GetBitContext gb;
00815     const ModeTab *mtab = tctx->mtab;
00816     float *out = data;
00817     enum FrameType ftype;
00818     int window_type;
00819     static const enum FrameType wtype_to_ftype_table[] = {
00820         FT_LONG,   FT_LONG, FT_SHORT, FT_LONG,
00821         FT_MEDIUM, FT_LONG, FT_LONG,  FT_MEDIUM, FT_MEDIUM
00822     };
00823 
00824     if (buf_size*8 < avctx->bit_rate*mtab->size/avctx->sample_rate + 8) {
00825         av_log(avctx, AV_LOG_ERROR,
00826                "Frame too small (%d bytes). Truncated file?\n", buf_size);
00827         *data_size = 0;
00828         return buf_size;
00829     }
00830 
00831     init_get_bits(&gb, buf, buf_size * 8);
00832     skip_bits(&gb, get_bits(&gb, 8));
00833     window_type = get_bits(&gb, WINDOW_TYPE_BITS);
00834 
00835     if (window_type > 8) {
00836         av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
00837         return -1;
00838     }
00839 
00840     ftype = wtype_to_ftype_table[window_type];
00841 
00842     read_and_decode_spectrum(tctx, &gb, tctx->spectrum, ftype);
00843 
00844     imdct_output(tctx, ftype, window_type, out);
00845 
00846     FFSWAP(float*, tctx->curr_frame, tctx->prev_frame);
00847 
00848     if (tctx->avctx->frame_number < 2) {
00849         *data_size=0;
00850         return buf_size;
00851     }
00852 
00853     *data_size = mtab->size*avctx->channels*4;
00854 
00855     return buf_size;
00856 }
00857 
00861 static av_cold void init_mdct_win(TwinContext *tctx)
00862 {
00863     int i,j;
00864     const ModeTab *mtab = tctx->mtab;
00865     int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;
00866     int size_m = mtab->size / mtab->fmode[FT_MEDIUM].sub;
00867     int channels = tctx->avctx->channels;
00868     float norm = channels == 1 ? 2. : 1.;
00869 
00870     for (i = 0; i < 3; i++) {
00871         int bsize = tctx->mtab->size/tctx->mtab->fmode[i].sub;
00872         ff_mdct_init(&tctx->mdct_ctx[i], av_log2(bsize) + 1, 1,
00873                      -sqrt(norm/bsize) / (1<<15));
00874     }
00875 
00876     tctx->tmp_buf  = av_malloc(mtab->size            * sizeof(*tctx->tmp_buf));
00877 
00878     tctx->spectrum  = av_malloc(2*mtab->size*channels*sizeof(float));
00879     tctx->curr_frame = av_malloc(2*mtab->size*channels*sizeof(float));
00880     tctx->prev_frame  = av_malloc(2*mtab->size*channels*sizeof(float));
00881 
00882     for (i = 0; i < 3; i++) {
00883         int m = 4*mtab->size/mtab->fmode[i].sub;
00884         double freq = 2*M_PI/m;
00885         tctx->cos_tabs[i] = av_malloc((m/4)*sizeof(*tctx->cos_tabs));
00886 
00887         for (j = 0; j <= m/8; j++)
00888             tctx->cos_tabs[i][j] = cos((2*j + 1)*freq);
00889         for (j = 1; j <  m/8; j++)
00890             tctx->cos_tabs[i][m/4-j] = tctx->cos_tabs[i][j];
00891     }
00892 
00893 
00894     ff_init_ff_sine_windows(av_log2(size_m));
00895     ff_init_ff_sine_windows(av_log2(size_s/2));
00896     ff_init_ff_sine_windows(av_log2(mtab->size));
00897 }
00898 
00905 static void permutate_in_line(int16_t *tab, int num_vect, int num_blocks,
00906                               int block_size,
00907                               const uint8_t line_len[2], int length_div,
00908                               enum FrameType ftype)
00909 
00910 {
00911     int i,j;
00912 
00913     for (i = 0; i < line_len[0]; i++) {
00914         int shift;
00915 
00916         if (num_blocks == 1 ||
00917             (ftype == FT_LONG && num_vect % num_blocks) ||
00918             (ftype != FT_LONG && num_vect & 1         ) ||
00919             i == line_len[1]) {
00920             shift = 0;
00921         } else if (ftype == FT_LONG) {
00922             shift = i;
00923         } else
00924             shift = i*i;
00925 
00926         for (j = 0; j < num_vect && (j+num_vect*i < block_size*num_blocks); j++)
00927             tab[i*num_vect+j] = i*num_vect + (j + shift) % num_vect;
00928     }
00929 }
00930 
00946 static void transpose_perm(int16_t *out, int16_t *in, int num_vect,
00947                            const uint8_t line_len[2], int length_div)
00948 {
00949     int i,j;
00950     int cont= 0;
00951     for (i = 0; i < num_vect; i++)
00952         for (j = 0; j < line_len[i >= length_div]; j++)
00953             out[cont++] = in[j*num_vect + i];
00954 }
00955 
00956 static void linear_perm(int16_t *out, int16_t *in, int n_blocks, int size)
00957 {
00958     int block_size = size/n_blocks;
00959     int i;
00960 
00961     for (i = 0; i < size; i++)
00962         out[i] = block_size * (in[i] % n_blocks) + in[i] / n_blocks;
00963 }
00964 
00965 static av_cold void construct_perm_table(TwinContext *tctx,enum FrameType ftype)
00966 {
00967     int block_size;
00968     const ModeTab *mtab = tctx->mtab;
00969     int size = tctx->avctx->channels*mtab->fmode[ftype].sub;
00970     int16_t *tmp_perm = (int16_t *) tctx->tmp_buf;
00971 
00972     if (ftype == FT_PPC) {
00973         size  = tctx->avctx->channels;
00974         block_size = mtab->ppc_shape_len;
00975     } else
00976         block_size = mtab->size / mtab->fmode[ftype].sub;
00977 
00978     permutate_in_line(tmp_perm, tctx->n_div[ftype], size,
00979                       block_size, tctx->length[ftype],
00980                       tctx->length_change[ftype], ftype);
00981 
00982     transpose_perm(tctx->permut[ftype], tmp_perm, tctx->n_div[ftype],
00983                    tctx->length[ftype], tctx->length_change[ftype]);
00984 
00985     linear_perm(tctx->permut[ftype], tctx->permut[ftype], size,
00986                 size*block_size);
00987 }
00988 
00989 static av_cold void init_bitstream_params(TwinContext *tctx)
00990 {
00991     const ModeTab *mtab = tctx->mtab;
00992     int n_ch = tctx->avctx->channels;
00993     int total_fr_bits = tctx->avctx->bit_rate*mtab->size/
00994                              tctx->avctx->sample_rate;
00995 
00996     int lsp_bits_per_block = n_ch*(mtab->lsp_bit0 + mtab->lsp_bit1 +
00997                                    mtab->lsp_split*mtab->lsp_bit2);
00998 
00999     int ppc_bits = n_ch*(mtab->pgain_bit + mtab->ppc_shape_bit +
01000                          mtab->ppc_period_bit);
01001 
01002     int bsize_no_main_cb[3];
01003     int bse_bits[3];
01004     int i;
01005     enum FrameType frametype;
01006 
01007     for (i = 0; i < 3; i++)
01008         // +1 for history usage switch
01009         bse_bits[i] = n_ch *
01010             (mtab->fmode[i].bark_n_coef * mtab->fmode[i].bark_n_bit + 1);
01011 
01012     bsize_no_main_cb[2] = bse_bits[2] + lsp_bits_per_block + ppc_bits +
01013                           WINDOW_TYPE_BITS + n_ch*GAIN_BITS;
01014 
01015     for (i = 0; i < 2; i++)
01016         bsize_no_main_cb[i] =
01017             lsp_bits_per_block + n_ch*GAIN_BITS + WINDOW_TYPE_BITS +
01018             mtab->fmode[i].sub*(bse_bits[i] + n_ch*SUB_GAIN_BITS);
01019 
01020     // The remaining bits are all used for the main spectrum coefficients
01021     for (i = 0; i < 4; i++) {
01022         int bit_size;
01023         int vect_size;
01024         int rounded_up, rounded_down, num_rounded_down, num_rounded_up;
01025         if (i == 3) {
01026             bit_size  = n_ch * mtab->ppc_shape_bit;
01027             vect_size = n_ch * mtab->ppc_shape_len;
01028         } else {
01029             bit_size = total_fr_bits - bsize_no_main_cb[i];
01030             vect_size = n_ch * mtab->size;
01031         }
01032 
01033         tctx->n_div[i] = (bit_size + 13) / 14;
01034 
01035         rounded_up   = (bit_size + tctx->n_div[i] - 1)/tctx->n_div[i];
01036         rounded_down = (bit_size           )/tctx->n_div[i];
01037         num_rounded_down = rounded_up * tctx->n_div[i] - bit_size;
01038         num_rounded_up = tctx->n_div[i] - num_rounded_down;
01039         tctx->bits_main_spec[0][i][0] = (rounded_up   + 1)/2;
01040         tctx->bits_main_spec[1][i][0] = (rounded_up      )/2;
01041         tctx->bits_main_spec[0][i][1] = (rounded_down + 1)/2;
01042         tctx->bits_main_spec[1][i][1] = (rounded_down    )/2;
01043         tctx->bits_main_spec_change[i] = num_rounded_up;
01044 
01045         rounded_up   = (vect_size + tctx->n_div[i] - 1)/tctx->n_div[i];
01046         rounded_down = (vect_size                     )/tctx->n_div[i];
01047         num_rounded_down = rounded_up * tctx->n_div[i] - vect_size;
01048         num_rounded_up = tctx->n_div[i] - num_rounded_down;
01049         tctx->length[i][0] = rounded_up;
01050         tctx->length[i][1] = rounded_down;
01051         tctx->length_change[i] = num_rounded_up;
01052     }
01053 
01054     for (frametype = FT_SHORT; frametype <= FT_PPC; frametype++)
01055         construct_perm_table(tctx, frametype);
01056 }
01057 
01058 static av_cold int twin_decode_init(AVCodecContext *avctx)
01059 {
01060     TwinContext *tctx = avctx->priv_data;
01061     int isampf = avctx->sample_rate/1000;
01062     int ibps = avctx->bit_rate/(1000 * avctx->channels);
01063 
01064     tctx->avctx       = avctx;
01065     avctx->sample_fmt = SAMPLE_FMT_FLT;
01066 
01067     if (avctx->channels > 2) {
01068         av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n",
01069                avctx->channels);
01070         return -1;
01071     }
01072 
01073     switch ((isampf << 8) +  ibps) {
01074     case (8 <<8) +  8: tctx->mtab = &mode_08_08; break;
01075     case (11<<8) +  8: tctx->mtab = &mode_11_08; break;
01076     case (11<<8) + 10: tctx->mtab = &mode_11_10; break;
01077     case (16<<8) + 16: tctx->mtab = &mode_16_16; break;
01078     case (22<<8) + 20: tctx->mtab = &mode_22_20; break;
01079     case (22<<8) + 24: tctx->mtab = &mode_22_24; break;
01080     case (22<<8) + 32: tctx->mtab = &mode_22_32; break;
01081     case (44<<8) + 40: tctx->mtab = &mode_44_40; break;
01082     case (44<<8) + 48: tctx->mtab = &mode_44_48; break;
01083     default:
01084         av_log(avctx, AV_LOG_ERROR, "This version does not support %d kHz - %d kbit/s/ch mode.\n", isampf, isampf);
01085         return -1;
01086     }
01087 
01088     dsputil_init(&tctx->dsp, avctx);
01089     init_mdct_win(tctx);
01090     init_bitstream_params(tctx);
01091 
01092     memset_float(tctx->bark_hist[0][0], 0.1, FF_ARRAY_ELEMS(tctx->bark_hist));
01093 
01094     return 0;
01095 }
01096 
01097 static av_cold int twin_decode_close(AVCodecContext *avctx)
01098 {
01099     TwinContext *tctx = avctx->priv_data;
01100     int i;
01101 
01102     for (i = 0; i < 3; i++) {
01103         ff_mdct_end(&tctx->mdct_ctx[i]);
01104         av_free(tctx->cos_tabs[i]);
01105     }
01106 
01107 
01108     av_free(tctx->curr_frame);
01109     av_free(tctx->spectrum);
01110     av_free(tctx->prev_frame);
01111     av_free(tctx->tmp_buf);
01112 
01113     return 0;
01114 }
01115 
01116 AVCodec twinvq_decoder =
01117 {
01118     "twinvq",
01119     AVMEDIA_TYPE_AUDIO,
01120     CODEC_ID_TWINVQ,
01121     sizeof(TwinContext),
01122     twin_decode_init,
01123     NULL,
01124     twin_decode_close,
01125     twin_decode_frame,
01126     .long_name = NULL_IF_CONFIG_SMALL("VQF TwinVQ"),
01127 };

Generated on Fri Sep 16 2011 17:17:44 for FFmpeg by  doxygen 1.7.1