1 /* ------------------------------------------------------------------ 2 * Copyright (C) 1998-2009 PacketVideo 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either 13 * express or implied. 14 * See the License for the specific language governing permissions 15 * and limitations under the License. 16 * ------------------------------------------------------------------- 17 */ 18 /**************************************************************************************** 19 Portions of this file are derived from the following 3GPP standard: 20 21 3GPP TS 26.073 22 ANSI-C code for the Adaptive Multi-Rate (AMR) speech codec 23 Available from http://www.3gpp.org 24 25 (C) 2004, 3GPP Organizational Partners (ARIB, ATIS, CCSA, ETSI, TTA, TTC) 26 Permission to distribute, modify and use this file under the standard license 27 terms listed above has been obtained from the copyright holder. 28 ****************************************************************************************/ 29 /* 30 ------------------------------------------------------------------------------ 31 32 33 34 Pathname: ./audio/gsm-amr/c/src/dtx_dec.c 35 Functions: 36 dtx_dec_reset 37 dtx_dec 38 dtx_dec_activity_update 39 rx_dtx_handler 40 41 ------------------------------------------------------------------------------ 42 MODULE DESCRIPTION 43 44 These modules decode the comfort noise when in DTX. 45 46 ------------------------------------------------------------------------------ 47 */ 48 49 50 /*---------------------------------------------------------------------------- 51 ; INCLUDES 52 ----------------------------------------------------------------------------*/ 53 #include <string.h> 54 55 #include "dtx_dec.h" 56 #include "typedef.h" 57 #include "basic_op.h" 58 #include "copy.h" 59 #include "set_zero.h" 60 #include "mode.h" 61 #include "log2.h" 62 #include "lsp_az.h" 63 #include "pow2.h" 64 #include "a_refl.h" 65 #include "b_cn_cod.h" 66 #include "syn_filt.h" 67 #include "lsp_lsf.h" 68 #include "reorder.h" 69 #include "lsp_tab.h" 70 71 /*---------------------------------------------------------------------------- 72 ; MACROS 73 ; Define module specific macros here 74 ----------------------------------------------------------------------------*/ 75 76 77 /*---------------------------------------------------------------------------- 78 ; DEFINES 79 ; Include all pre-processor statements here. Include conditional 80 ; compile variables also. 81 ----------------------------------------------------------------------------*/ 82 #define PN_INITIAL_SEED 0x70816958L /* Pseudo noise generator seed value */ 83 84 /*---------------------------------------------------------------------------- 85 ; LOCAL FUNCTION DEFINITIONS 86 ; Function Prototype declaration 87 ----------------------------------------------------------------------------*/ 88 89 90 /*---------------------------------------------------------------------------- 91 ; LOCAL VARIABLE DEFINITIONS 92 ; Variable declaration - defined here and used outside this module 93 ----------------------------------------------------------------------------*/ 94 95 /*************************************************** 96 * Scaling factors for the lsp variability operation * 97 ***************************************************/ 98 static const Word16 lsf_hist_mean_scale[M] = 99 { 100 20000, 101 20000, 102 20000, 103 20000, 104 20000, 105 18000, 106 16384, 107 8192, 108 0, 109 0 110 }; 111 112 /************************************************* 113 * level adjustment for different modes Q11 * 114 *************************************************/ 115 static const Word16 dtx_log_en_adjust[9] = 116 { 117 -1023, /* MR475 */ 118 -878, /* MR515 */ 119 -732, /* MR59 */ 120 -586, /* MR67 */ 121 -440, /* MR74 */ 122 -294, /* MR795 */ 123 -148, /* MR102 */ 124 0, /* MR122 */ 125 0, /* MRDTX */ 126 }; 127 128 129 /* 130 ------------------------------------------------------------------------------ 131 FUNCTION NAME: dtx_dec_reset 132 ------------------------------------------------------------------------------ 133 INPUT AND OUTPUT DEFINITIONS 134 135 Inputs: 136 st = pointer to a structure of type dtx_decState 137 138 Outputs: 139 Structure pointed to by st is initialized to a set of initial values. 140 141 Returns: 142 return_value = 0 if memory was successfully initialized, 143 otherwise returns -1 (int) 144 145 Global Variables Used: 146 None. 147 148 Local Variables Needed: 149 None. 150 151 ------------------------------------------------------------------------------ 152 FUNCTION DESCRIPTION 153 154 Reset of state memory for dtx_dec. 155 156 ------------------------------------------------------------------------------ 157 REQUIREMENTS 158 159 None. 160 161 ------------------------------------------------------------------------------ 162 REFERENCES 163 164 dtx_dec.c, UMTS GSM AMR speech codec, R99 - Version 3.3.0, December 12, 2001 165 166 ------------------------------------------------------------------------------ 167 PSEUDO-CODE 168 169 int dtx_dec_reset (dtx_decState *st) 170 { 171 int i; 172 173 if (st == (dtx_decState *) NULL){ 174 fprintf(stderr, "dtx_dec_reset: invalid parameter\n"); 175 return -1; 176 } 177 178 st->since_last_sid = 0; 179 st->true_sid_period_inv = (1 << 13); 180 181 st->log_en = 3500; 182 st->old_log_en = 3500; 183 // low level noise for better performance in DTX handover cases 184 185 st->L_pn_seed_rx = PN_INITIAL_SEED; 186 187 // Initialize state->lsp [] and state->lsp_old [] 188 Copy(lsp_init_data, &st->lsp[0], M); 189 Copy(lsp_init_data, &st->lsp_old[0], M); 190 191 st->lsf_hist_ptr = 0; 192 st->log_pg_mean = 0; 193 st->log_en_hist_ptr = 0; 194 195 // initialize decoder lsf history 196 Copy(mean_lsf, &st->lsf_hist[0], M); 197 198 for (i = 1; i < DTX_HIST_SIZE; i++) 199 { 200 Copy(&st->lsf_hist[0], &st->lsf_hist[M*i], M); 201 } 202 Set_zero(st->lsf_hist_mean, M*DTX_HIST_SIZE); 203 204 // initialize decoder log frame energy 205 for (i = 0; i < DTX_HIST_SIZE; i++) 206 { 207 st->log_en_hist[i] = st->log_en; 208 } 209 210 st->log_en_adjust = 0; 211 212 st->dtxHangoverCount = DTX_HANG_CONST; 213 st->decAnaElapsedCount = 32767; 214 215 st->sid_frame = 0; 216 st->valid_data = 0; 217 st->dtxHangoverAdded = 0; 218 219 st->dtxGlobalState = DTX; 220 st->data_updated = 0; 221 return 0; 222 } 223 224 ------------------------------------------------------------------------------ 225 RESOURCES USED [optional] 226 227 When the code is written for a specific target processor the 228 the resources used should be documented below. 229 230 HEAP MEMORY USED: x bytes 231 232 STACK MEMORY USED: x bytes 233 234 CLOCK CYCLES: (cycle count equation for this function) + (variable 235 used to represent cycle count for each subroutine 236 called) 237 where: (cycle count variable) = cycle count for [subroutine 238 name] 239 240 ------------------------------------------------------------------------------ 241 CAUTION [optional] 242 [State any special notes, constraints or cautions for users of this function] 243 244 ------------------------------------------------------------------------------ 245 */ 246 247 Word16 dtx_dec_reset(dtx_decState *st) 248 { 249 Word16 i; 250 251 if (st == (dtx_decState *) NULL) 252 { 253 /* fprint(stderr, "dtx_dec_reset: invalid parameter\n"); */ 254 return(-1); 255 } 256 257 st->since_last_sid = 0; 258 st->true_sid_period_inv = (1 << 13); 259 260 st->log_en = 3500; 261 st->old_log_en = 3500; 262 /* low level noise for better performance in DTX handover cases*/ 263 264 st->L_pn_seed_rx = PN_INITIAL_SEED; 265 266 /* Initialize state->lsp [] */ 267 st->lsp[0] = 30000; 268 st->lsp[1] = 26000; 269 st->lsp[2] = 21000; 270 st->lsp[3] = 15000; 271 st->lsp[4] = 8000; 272 st->lsp[5] = 0; 273 st->lsp[6] = -8000; 274 st->lsp[7] = -15000; 275 st->lsp[8] = -21000; 276 st->lsp[9] = -26000; 277 278 /* Initialize state->lsp_old [] */ 279 st->lsp_old[0] = 30000; 280 st->lsp_old[1] = 26000; 281 st->lsp_old[2] = 21000; 282 st->lsp_old[3] = 15000; 283 st->lsp_old[4] = 8000; 284 st->lsp_old[5] = 0; 285 st->lsp_old[6] = -8000; 286 st->lsp_old[7] = -15000; 287 st->lsp_old[8] = -21000; 288 st->lsp_old[9] = -26000; 289 290 st->lsf_hist_ptr = 0; 291 st->log_pg_mean = 0; 292 st->log_en_hist_ptr = 0; 293 294 /* initialize decoder lsf history */ 295 st->lsf_hist[0] = 1384; 296 st->lsf_hist[1] = 2077; 297 st->lsf_hist[2] = 3420; 298 st->lsf_hist[3] = 5108; 299 st->lsf_hist[4] = 6742; 300 st->lsf_hist[5] = 8122; 301 st->lsf_hist[6] = 9863; 302 st->lsf_hist[7] = 11092; 303 st->lsf_hist[8] = 12714; 304 st->lsf_hist[9] = 13701; 305 306 for (i = 1; i < DTX_HIST_SIZE; i++) 307 { 308 Copy(&st->lsf_hist[0], &st->lsf_hist[M*i], M); 309 } 310 memset(st->lsf_hist_mean, 0, sizeof(Word16)*M*DTX_HIST_SIZE); 311 312 /* initialize decoder log frame energy */ 313 for (i = 0; i < DTX_HIST_SIZE; i++) 314 { 315 st->log_en_hist[i] = st->log_en; 316 } 317 318 st->log_en_adjust = 0; 319 320 st->dtxHangoverCount = DTX_HANG_CONST; 321 st->decAnaElapsedCount = 32767; 322 323 st->sid_frame = 0; 324 st->valid_data = 0; 325 st->dtxHangoverAdded = 0; 326 327 st->dtxGlobalState = DTX; 328 st->data_updated = 0; 329 330 return(0); 331 } 332 333 /****************************************************************************/ 334 335 /* 336 ------------------------------------------------------------------------------ 337 FUNCTION NAME: dtx_dec 338 ------------------------------------------------------------------------------ 339 INPUT AND OUTPUT DEFINITIONS 340 341 Inputs: 342 st = pointer to a structure of type dtx_decState 343 mem_syn = AMR decoder state 344 lsfState = decoder lsf states 345 predState = prediction states 346 averState = CB gain average states 347 new_state = new DTX state 348 mode = AMR mode 349 parm = Vector of synthesis parameters 350 351 Outputs: 352 st points to an updated structure of type dtx_decState 353 mem_syn = AMR decoder state 354 lsfState = decoder lsf states 355 predState = prediction states 356 averState = CB gain average states 357 synth = synthesised speech 358 A_t = decoded LP filter in 4 subframes 359 360 Returns: 361 return_value = 0 (int) 362 363 Global Variables Used: 364 None. 365 366 Local Variables Needed: 367 None. 368 369 ------------------------------------------------------------------------------ 370 FUNCTION DESCRIPTION 371 372 Decode comfort noise when in DTX. 373 374 ------------------------------------------------------------------------------ 375 REQUIREMENTS 376 377 None 378 379 ------------------------------------------------------------------------------ 380 REFERENCES 381 382 dtx_dec.c, UMTS GSM AMR speech codec, R99 - Version 3.3.0, December 12, 2001 383 384 ------------------------------------------------------------------------------ 385 PSEUDO-CODE 386 387 int dtx_dec( 388 dtx_decState *st, // i/o : State struct 389 Word16 mem_syn[], // i/o : AMR decoder state 390 D_plsfState* lsfState, // i/o : decoder lsf states 391 gc_predState* predState, // i/o : prediction states 392 Cb_gain_averageState* averState, // i/o : CB gain average states 393 enum DTXStateType new_state, // i : new DTX state 394 enum Mode mode, // i : AMR mode 395 Word16 parm[], // i : Vector of synthesis parameters 396 Word16 synth[], // o : synthesised speech 397 Word16 A_t[] // o : decoded LP filter in 4 subframes 398 ) 399 { 400 Word16 log_en_index; 401 Word16 i, j; 402 Word16 int_fac; 403 Word32 L_log_en_int; 404 Word16 lsp_int[M]; 405 Word16 log_en_int_e; 406 Word16 log_en_int_m; 407 Word16 level; 408 Word16 acoeff[M + 1]; 409 Word16 refl[M]; 410 Word16 pred_err; 411 Word16 ex[L_SUBFR]; 412 Word16 ma_pred_init; 413 Word16 log_pg_e, log_pg_m; 414 Word16 log_pg; 415 Flag negative; 416 Word16 lsf_mean; 417 Word32 L_lsf_mean; 418 Word16 lsf_variab_index; 419 Word16 lsf_variab_factor; 420 Word16 lsf_int[M]; 421 Word16 lsf_int_variab[M]; 422 Word16 lsp_int_variab[M]; 423 Word16 acoeff_variab[M + 1]; 424 425 Word16 lsf[M]; 426 Word32 L_lsf[M]; 427 Word16 ptr; 428 Word16 tmp_int_length; 429 430 431 // This function is called if synthesis state is not SPEECH 432 // the globally passed inputs to this function are 433 // st->sid_frame 434 // st->valid_data 435 // st->dtxHangoverAdded 436 // new_state (SPEECH, DTX, DTX_MUTE) 437 438 if ((st->dtxHangoverAdded != 0) && 439 (st->sid_frame != 0)) 440 { 441 // sid_first after dtx hangover period 442 // or sid_upd after dtxhangover 443 444 // set log_en_adjust to correct value 445 st->log_en_adjust = dtx_log_en_adjust[mode]; 446 447 ptr = add(st->lsf_hist_ptr, M); 448 if (sub(ptr, 80) == 0) 449 { 450 ptr = 0; 451 } 452 Copy( &st->lsf_hist[st->lsf_hist_ptr],&st->lsf_hist[ptr],M); 453 454 ptr = add(st->log_en_hist_ptr,1); 455 if (sub(ptr, DTX_HIST_SIZE) == 0) 456 { 457 ptr = 0; 458 } 459 st->log_en_hist[ptr] = st->log_en_hist[st->log_en_hist_ptr]; // Q11 460 461 // compute mean log energy and lsp 462 // from decoded signal (SID_FIRST) 463 st->log_en = 0; 464 for (i = 0; i < M; i++) 465 { 466 L_lsf[i] = 0; 467 } 468 469 // average energy and lsp 470 for (i = 0; i < DTX_HIST_SIZE; i++) 471 { 472 st->log_en = add(st->log_en, 473 shr(st->log_en_hist[i],3)); 474 for (j = 0; j < M; j++) 475 { 476 L_lsf[j] = L_add(L_lsf[j], 477 L_deposit_l(st->lsf_hist[i * M + j])); 478 } 479 } 480 481 for (j = 0; j < M; j++) 482 { 483 lsf[j] = extract_l(L_shr(L_lsf[j],3)); // divide by 8 484 } 485 486 Lsf_lsp(lsf, st->lsp, M); 487 488 // make log_en speech coder mode independent 489 // added again later before synthesis 490 st->log_en = sub(st->log_en, st->log_en_adjust); 491 492 // compute lsf variability vector 493 Copy(st->lsf_hist, st->lsf_hist_mean, 80); 494 495 for (i = 0; i < M; i++) 496 { 497 L_lsf_mean = 0; 498 // compute mean lsf 499 for (j = 0; j < 8; j++) 500 { 501 L_lsf_mean = L_add(L_lsf_mean, 502 L_deposit_l(st->lsf_hist_mean[i+j*M])); 503 } 504 505 lsf_mean = extract_l(L_shr(L_lsf_mean, 3)); 506 // subtract mean and limit to within reasonable limits 507 // moreover the upper lsf's are attenuated 508 for (j = 0; j < 8; j++) 509 { 510 // subtract mean 511 st->lsf_hist_mean[i+j*M] = 512 sub(st->lsf_hist_mean[i+j*M], lsf_mean); 513 514 // attenuate deviation from mean, especially for upper lsf's 515 st->lsf_hist_mean[i+j*M] = 516 mult(st->lsf_hist_mean[i+j*M], lsf_hist_mean_scale[i]); 517 518 // limit the deviation 519 if (st->lsf_hist_mean[i+j*M] < 0) 520 { 521 negative = 1; 522 } 523 else 524 { 525 negative = 0; 526 } 527 st->lsf_hist_mean[i+j*M] = abs_s(st->lsf_hist_mean[i+j*M]); 528 529 // apply soft limit 530 if (sub(st->lsf_hist_mean[i+j*M], 655) > 0) 531 { 532 st->lsf_hist_mean[i+j*M] = 533 add(655, shr(sub(st->lsf_hist_mean[i+j*M], 655), 2)); 534 } 535 536 // apply hard limit 537 if (sub(st->lsf_hist_mean[i+j*M], 1310) > 0) 538 { 539 st->lsf_hist_mean[i+j*M] = 1310; 540 } 541 if (negative != 0) 542 { 543 st->lsf_hist_mean[i+j*M] = -st->lsf_hist_mean[i+j*M]; 544 } 545 546 } 547 } 548 } 549 550 if (st->sid_frame != 0 ) 551 { 552 // Set old SID parameters, always shift 553 // even if there is no new valid_data 554 Copy(st->lsp, st->lsp_old, M); 555 st->old_log_en = st->log_en; 556 557 if (st->valid_data != 0 ) // new data available (no CRC) 558 { 559 // Compute interpolation factor, since the division only works 560 // for values of since_last_sid < 32 we have to limit the 561 // interpolation to 32 frames 562 tmp_int_length = st->since_last_sid; 563 st->since_last_sid = 0; 564 565 if (sub(tmp_int_length, 32) > 0) 566 { 567 tmp_int_length = 32; 568 } 569 if (sub(tmp_int_length, 2) >= 0) 570 { 571 st->true_sid_period_inv = div_s(1 << 10, shl(tmp_int_length, 10)); 572 } 573 else 574 { 575 st->true_sid_period_inv = 1 << 14; // 0.5 it Q15 576 } 577 578 Init_D_plsf_3(lsfState, parm[0]); // temporay initialization 579 D_plsf_3(lsfState, MRDTX, 0, &parm[1], st->lsp); 580 Set_zero(lsfState->past_r_q, M); // reset for next speech frame 581 582 log_en_index = parm[4]; 583 // Q11 and divide by 4 584 st->log_en = shl(log_en_index, (11 - 2)); 585 586 // Subtract 2.5 in Q11 587 st->log_en = sub(st->log_en, (2560 * 2)); 588 589 // Index 0 is reserved for silence 590 if (log_en_index == 0) 591 { 592 st->log_en = MIN_16; 593 } 594 595 // no interpolation at startup after coder reset 596 // or when SID_UPD has been received right after SPEECH 597 if ((st->data_updated == 0) || 598 (sub(st->dtxGlobalState, SPEECH) == 0) 599 ) 600 { 601 Copy(st->lsp, st->lsp_old, M); 602 st->old_log_en = st->log_en; 603 } 604 } // endif valid_data 605 606 // initialize gain predictor memory of other modes 607 ma_pred_init = sub(shr(st->log_en,1), 9000); 608 if (ma_pred_init > 0) 609 { 610 ma_pred_init = 0; 611 } 612 if (sub(ma_pred_init, -14436) < 0) 613 { 614 ma_pred_init = -14436; 615 } 616 617 predState->past_qua_en[0] = ma_pred_init; 618 predState->past_qua_en[1] = ma_pred_init; 619 predState->past_qua_en[2] = ma_pred_init; 620 predState->past_qua_en[3] = ma_pred_init; 621 622 // past_qua_en for other modes than MR122 623 ma_pred_init = mult(5443, ma_pred_init); 624 // scale down by factor 20*log10(2) in Q15 625 predState->past_qua_en_MR122[0] = ma_pred_init; 626 predState->past_qua_en_MR122[1] = ma_pred_init; 627 predState->past_qua_en_MR122[2] = ma_pred_init; 628 predState->past_qua_en_MR122[3] = ma_pred_init; 629 } // endif sid_frame 630 631 // CN generation 632 // recompute level adjustment factor Q11 633 // st->log_en_adjust = 0.9*st->log_en_adjust + 634 // 0.1*dtx_log_en_adjust[mode]); 635 st->log_en_adjust = add(mult(st->log_en_adjust, 29491), 636 shr(mult(shl(dtx_log_en_adjust[mode],5),3277),5)); 637 638 // Interpolate SID info 639 int_fac = shl(add(1,st->since_last_sid), 10); // Q10 640 int_fac = mult(int_fac, st->true_sid_period_inv); // Q10 * Q15 -> Q10 641 642 // Maximize to 1.0 in Q10 643 if (sub(int_fac, 1024) > 0) 644 { 645 int_fac = 1024; 646 } 647 int_fac = shl(int_fac, 4); // Q10 -> Q14 648 649 L_log_en_int = L_mult(int_fac, st->log_en); // Q14 * Q11->Q26 650 for(i = 0; i < M; i++) 651 { 652 lsp_int[i] = mult(int_fac, st->lsp[i]);// Q14 * Q15 -> Q14 653 } 654 655 int_fac = sub(16384, int_fac); // 1-k in Q14 656 657 // (Q14 * Q11 -> Q26) + Q26 -> Q26 658 L_log_en_int = L_mac(L_log_en_int, int_fac, st->old_log_en); 659 for(i = 0; i < M; i++) 660 { 661 // Q14 + (Q14 * Q15 -> Q14) -> Q14 662 lsp_int[i] = add(lsp_int[i], mult(int_fac, st->lsp_old[i])); 663 lsp_int[i] = shl(lsp_int[i], 1); // Q14 -> Q15 664 } 665 666 // compute the amount of lsf variability 667 lsf_variab_factor = sub(st->log_pg_mean,2457); // -0.6 in Q12 668 // *0.3 Q12*Q15 -> Q12 669 lsf_variab_factor = sub(4096, mult(lsf_variab_factor, 9830)); 670 671 // limit to values between 0..1 in Q12 672 if (sub(lsf_variab_factor, 4096) > 0) 673 { 674 lsf_variab_factor = 4096; 675 } 676 if (lsf_variab_factor < 0) 677 { 678 lsf_variab_factor = 0; 679 } 680 lsf_variab_factor = shl(lsf_variab_factor, 3); // -> Q15 681 682 // get index of vector to do variability with 683 lsf_variab_index = pseudonoise(&st->L_pn_seed_rx, 3); 684 685 // convert to lsf 686 Lsp_lsf(lsp_int, lsf_int, M); 687 688 // apply lsf variability 689 Copy(lsf_int, lsf_int_variab, M); 690 for(i = 0; i < M; i++) 691 { 692 lsf_int_variab[i] = add(lsf_int_variab[i], 693 mult(lsf_variab_factor, 694 st->lsf_hist_mean[i+lsf_variab_index*M])); 695 } 696 697 // make sure that LSP's are ordered 698 Reorder_lsf(lsf_int, LSF_GAP, M); 699 Reorder_lsf(lsf_int_variab, LSF_GAP, M); 700 701 // copy lsf to speech decoders lsf state 702 Copy(lsf_int, lsfState->past_lsf_q, M); 703 704 // convert to lsp 705 Lsf_lsp(lsf_int, lsp_int, M); 706 Lsf_lsp(lsf_int_variab, lsp_int_variab, M); 707 708 // Compute acoeffs Q12 acoeff is used for level 709 // normalization and postfilter, acoeff_variab is 710 // used for synthesis filter 711 // by doing this we make sure that the level 712 // in high frequenncies does not jump up and down 713 714 Lsp_Az(lsp_int, acoeff); 715 Lsp_Az(lsp_int_variab, acoeff_variab); 716 717 // For use in postfilter 718 Copy(acoeff, &A_t[0], M + 1); 719 Copy(acoeff, &A_t[M + 1], M + 1); 720 Copy(acoeff, &A_t[2 * (M + 1)], M + 1); 721 Copy(acoeff, &A_t[3 * (M + 1)], M + 1); 722 723 // Compute reflection coefficients Q15 724 A_Refl(&acoeff[1], refl); 725 726 // Compute prediction error in Q15 727 pred_err = MAX_16; // 0.99997 in Q15 728 for (i = 0; i < M; i++) 729 { 730 pred_err = mult(pred_err, sub(MAX_16, mult(refl[i], refl[i]))); 731 } 732 733 // compute logarithm of prediction gain 734 Log2(L_deposit_l(pred_err), &log_pg_e, &log_pg_m); 735 736 // convert exponent and mantissa to Word16 Q12 737 log_pg = shl(sub(log_pg_e,15), 12); // Q12 738 log_pg = shr(sub(0,add(log_pg, shr(log_pg_m, 15-12))), 1); 739 st->log_pg_mean = add(mult(29491,st->log_pg_mean), 740 mult(3277, log_pg)); 741 742 // Compute interpolated log energy 743 L_log_en_int = L_shr(L_log_en_int, 10); // Q26 -> Q16 744 745 // Add 4 in Q16 746 L_log_en_int = L_add(L_log_en_int, 4 * 65536L); 747 748 // subtract prediction gain 749 L_log_en_int = L_sub(L_log_en_int, L_shl(L_deposit_l(log_pg), 4)); 750 751 // adjust level to speech coder mode 752 L_log_en_int = L_add(L_log_en_int, 753 L_shl(L_deposit_l(st->log_en_adjust), 5)); 754 755 log_en_int_e = extract_h(L_log_en_int); 756 log_en_int_m = extract_l(L_shr(L_sub(L_log_en_int, 757 L_deposit_h(log_en_int_e)), 1)); 758 level = extract_l(Pow2(log_en_int_e, log_en_int_m)); // Q4 759 760 for (i = 0; i < 4; i++) 761 { 762 // Compute innovation vector 763 build_CN_code(&st->L_pn_seed_rx, ex); 764 for (j = 0; j < L_SUBFR; j++) 765 { 766 ex[j] = mult(level, ex[j]); 767 } 768 // Synthesize 769 Syn_filt(acoeff_variab, ex, &synth[i * L_SUBFR], L_SUBFR, 770 mem_syn, 1); 771 772 } // next i 773 774 // reset codebook averaging variables 775 averState->hangVar = 20; 776 averState->hangCount = 0; 777 778 if (sub(new_state, DTX_MUTE) == 0) 779 { 780 // mute comfort noise as it has been quite a long time since 781 * last SID update was performed 782 783 tmp_int_length = st->since_last_sid; 784 if (sub(tmp_int_length, 32) > 0) 785 { 786 tmp_int_length = 32; 787 } 788 789 // safety guard against division by zero 790 if(tmp_int_length <= 0) { 791 tmp_int_length = 8; 792 } 793 794 st->true_sid_period_inv = div_s(1 << 10, shl(tmp_int_length, 10)); 795 796 st->since_last_sid = 0; 797 Copy(st->lsp, st->lsp_old, M); 798 st->old_log_en = st->log_en; 799 // subtract 1/8 in Q11 i.e -6/8 dB 800 st->log_en = sub(st->log_en, 256); 801 } 802 803 // reset interpolation length timer 804 // if data has been updated. 805 if ((st->sid_frame != 0) && 806 ((st->valid_data != 0) || 807 ((st->valid_data == 0) && (st->dtxHangoverAdded) != 0))) 808 { 809 st->since_last_sid = 0; 810 st->data_updated = 1; 811 } 812 813 return 0; 814 } 815 816 817 ------------------------------------------------------------------------------ 818 RESOURCES USED [optional] 819 820 When the code is written for a specific target processor the 821 the resources used should be documented below. 822 823 HEAP MEMORY USED: x bytes 824 825 STACK MEMORY USED: x bytes 826 827 CLOCK CYCLES: (cycle count equation for this function) + (variable 828 used to represent cycle count for each subroutine 829 called) 830 where: (cycle count variable) = cycle count for [subroutine 831 name] 832 833 ------------------------------------------------------------------------------ 834 CAUTION [optional] 835 [State any special notes, constraints or cautions for users of this function] 836 837 ------------------------------------------------------------------------------ 838 */ 839 840 void dtx_dec( 841 dtx_decState *st, /* i/o : State struct */ 842 Word16 mem_syn[], /* i/o : AMR decoder state */ 843 D_plsfState* lsfState, /* i/o : decoder lsf states */ 844 gc_predState* predState, /* i/o : prediction states */ 845 Cb_gain_averageState* averState, /* i/o : CB gain average states */ 846 enum DTXStateType new_state, /* i : new DTX state */ 847 enum Mode mode, /* i : AMR mode */ 848 Word16 parm[], /* i : Vector of synthesis parameters */ 849 Word16 synth[], /* o : synthesised speech */ 850 Word16 A_t[], /* o : decoded LP filter in 4 subframes*/ 851 Flag *pOverflow 852 ) 853 { 854 Word16 log_en_index; 855 Word16 i; 856 Word16 j; 857 Word16 int_fac; 858 Word32 L_log_en_int; 859 Word16 lsp_int[M]; 860 Word16 log_en_int_e; 861 Word16 log_en_int_m; 862 Word16 level; 863 Word16 acoeff[M + 1]; 864 Word16 refl[M]; 865 Word16 pred_err; 866 Word16 ex[L_SUBFR]; 867 Word16 ma_pred_init; 868 Word16 log_pg_e; 869 Word16 log_pg_m; 870 Word16 log_pg; 871 Flag negative; 872 Word16 lsf_mean; 873 Word32 L_lsf_mean; 874 Word16 lsf_variab_index; 875 Word16 lsf_variab_factor; 876 Word16 lsf_int[M]; 877 Word16 lsf_int_variab[M]; 878 Word16 lsp_int_variab[M]; 879 Word16 acoeff_variab[M + 1]; 880 881 Word16 lsf[M]; 882 Word32 L_lsf[M]; 883 Word16 ptr; 884 Word16 tmp_int_length; 885 886 Word32 L_temp; 887 Word16 temp; 888 889 /* This function is called if synthesis state is not SPEECH 890 * the globally passed inputs to this function are 891 * st->sid_frame 892 * st->valid_data 893 * st->dtxHangoverAdded 894 * new_state (SPEECH, DTX, DTX_MUTE) 895 */ 896 897 if ((st->dtxHangoverAdded != 0) && 898 (st->sid_frame != 0)) 899 { 900 /* sid_first after dtx hangover period */ 901 /* or sid_upd after dtxhangover */ 902 903 /* set log_en_adjust to correct value */ 904 st->log_en_adjust = dtx_log_en_adjust[mode]; 905 906 ptr = st->lsf_hist_ptr + M; 907 908 if (ptr == 80) 909 { 910 ptr = 0; 911 } 912 Copy(&st->lsf_hist[st->lsf_hist_ptr], &st->lsf_hist[ptr], M); 913 914 ptr = st->log_en_hist_ptr + 1; 915 916 if (ptr == DTX_HIST_SIZE) 917 { 918 ptr = 0; 919 } 920 921 st->log_en_hist[ptr] = st->log_en_hist[st->log_en_hist_ptr]; /* Q11 */ 922 923 /* compute mean log energy and lsp * 924 * from decoded signal (SID_FIRST) */ 925 st->log_en = 0; 926 for (i = M - 1; i >= 0; i--) 927 { 928 L_lsf[i] = 0; 929 } 930 931 /* average energy and lsp */ 932 for (i = DTX_HIST_SIZE - 1; i >= 0; i--) 933 { 934 if (st->log_en_hist[i] < 0) 935 { 936 temp = ~((~st->log_en_hist[i]) >> 3); 937 } 938 else 939 { 940 temp = st->log_en_hist[i] >> 3; 941 } 942 st->log_en = add(st->log_en, temp, pOverflow); 943 for (j = M - 1; j >= 0; j--) 944 { 945 L_lsf[j] = L_add(L_lsf[j], 946 L_deposit_l(st->lsf_hist[i * M + j]), pOverflow); 947 } 948 } 949 950 for (j = M - 1; j >= 0; j--) 951 { 952 if (L_lsf[j] < 0) 953 { 954 lsf[j] = (Word16)(~((~L_lsf[j]) >> 3)); 955 } 956 else 957 { 958 lsf[j] = (Word16)(L_lsf[j] >> 3); 959 } 960 } 961 962 Lsf_lsp(lsf, st->lsp, M, pOverflow); 963 964 /* make log_en speech coder mode independent */ 965 /* added again later before synthesis */ 966 st->log_en = sub(st->log_en, st->log_en_adjust, pOverflow); 967 968 /* compute lsf variability vector */ 969 Copy(st->lsf_hist, st->lsf_hist_mean, 80); 970 971 for (i = M - 1; i >= 0; i--) 972 { 973 L_lsf_mean = 0; 974 /* compute mean lsf */ 975 for (j = 8 - 1; j >= 0; j--) 976 { 977 L_lsf_mean = L_add(L_lsf_mean, 978 L_deposit_l(st->lsf_hist_mean[i+j*M]), pOverflow); 979 } 980 981 if (L_lsf_mean < 0) 982 { 983 lsf_mean = (Word16)(~((~L_lsf_mean) >> 3)); 984 } 985 else 986 { 987 lsf_mean = (Word16)(L_lsf_mean >> 3); 988 } 989 /* subtract mean and limit to within reasonable limits * 990 * moreover the upper lsf's are attenuated */ 991 for (j = 8 - 1; j >= 0; j--) 992 { 993 /* subtract mean */ 994 st->lsf_hist_mean[i+j*M] = 995 sub(st->lsf_hist_mean[i+j*M], lsf_mean, pOverflow); 996 997 /* attenuate deviation from mean, especially for upper lsf's */ 998 st->lsf_hist_mean[i+j*M] = 999 mult(st->lsf_hist_mean[i+j*M], lsf_hist_mean_scale[i], pOverflow); 1000 1001 /* limit the deviation */ 1002 if (st->lsf_hist_mean[i+j*M] < 0) 1003 { 1004 negative = 1; 1005 } 1006 else 1007 { 1008 negative = 0; 1009 } 1010 st->lsf_hist_mean[i+j*M] = abs_s(st->lsf_hist_mean[i+j*M]); 1011 1012 /* apply soft limit */ 1013 if (st->lsf_hist_mean[i+j*M] > 655) 1014 { 1015 st->lsf_hist_mean[i+j*M] = 655 + ((st->lsf_hist_mean[i+j*M] 1016 - 655) >> 2); 1017 } 1018 1019 /* apply hard limit */ 1020 if (st->lsf_hist_mean[i+j*M] > 1310) 1021 { 1022 st->lsf_hist_mean[i+j*M] = 1310; 1023 } 1024 1025 if (negative != 0) 1026 { 1027 st->lsf_hist_mean[i+j*M] = -st->lsf_hist_mean[i+j*M]; 1028 } 1029 } 1030 } 1031 } 1032 1033 1034 if (st->sid_frame != 0) 1035 { 1036 /* Set old SID parameters, always shift */ 1037 /* even if there is no new valid_data */ 1038 Copy(st->lsp, st->lsp_old, M); 1039 st->old_log_en = st->log_en; 1040 1041 if (st->valid_data != 0) /* new data available (no CRC) */ 1042 { 1043 /* Compute interpolation factor, since the division only works * 1044 * for values of since_last_sid < 32 we have to limit the * 1045 * interpolation to 32 frames */ 1046 tmp_int_length = st->since_last_sid; 1047 st->since_last_sid = 0; 1048 1049 if (tmp_int_length >= 32) 1050 { 1051 tmp_int_length = 32; 1052 } 1053 1054 L_temp = ((Word32) tmp_int_length) << 10; 1055 if (L_temp != (Word32)((Word16) L_temp)) 1056 { 1057 *pOverflow = 1; 1058 L_temp = (Word32)((tmp_int_length > 0) ? MAX_16 : MIN_16); 1059 } 1060 temp = (Word16) L_temp; 1061 1062 if (tmp_int_length >= 2) 1063 { 1064 st->true_sid_period_inv = div_s(1 << 10, temp); 1065 } 1066 else 1067 { 1068 st->true_sid_period_inv = 1 << 14; /* 0.5 it Q15 */ 1069 } 1070 1071 Init_D_plsf_3(lsfState, parm[0]); 1072 D_plsf_3(lsfState, MRDTX, 0, &parm[1], st->lsp, pOverflow); 1073 Set_zero(lsfState->past_r_q, M); /* reset for next speech frame */ 1074 1075 log_en_index = parm[4]; 1076 /* Q11 and divide by 4 */ 1077 if ((log_en_index > 63) || (log_en_index < -64)) 1078 { 1079 st->log_en = (log_en_index > 0) ? MAX_16 : MIN_16; 1080 } 1081 else 1082 { 1083 st->log_en = (log_en_index) << (11 - 2); 1084 } 1085 1086 /* Subtract 2.5 in Q11 */ 1087 st->log_en = sub(st->log_en, (2560 * 2), pOverflow); 1088 1089 /* Index 0 is reserved for silence */ 1090 if (log_en_index == 0) 1091 { 1092 st->log_en = MIN_16; 1093 } 1094 1095 /* no interpolation at startup after coder reset */ 1096 /* or when SID_UPD has been received right after SPEECH */ 1097 1098 if ((st->data_updated == 0) || 1099 (st->dtxGlobalState == SPEECH)) 1100 { 1101 Copy(st->lsp, st->lsp_old, M); 1102 st->old_log_en = st->log_en; 1103 } 1104 } /* endif valid_data */ 1105 1106 /* initialize gain predictor memory of other modes */ 1107 if (st->log_en < 0) 1108 { 1109 temp = ~((~st->log_en) >> 1); 1110 } 1111 else 1112 { 1113 temp = st->log_en >> 1; 1114 } 1115 ma_pred_init = sub(temp, 9000, pOverflow); 1116 1117 if (ma_pred_init > 0) 1118 { 1119 ma_pred_init = 0; 1120 } 1121 else if (ma_pred_init < -14436) 1122 { 1123 ma_pred_init = -14436; 1124 } 1125 1126 predState->past_qua_en[0] = ma_pred_init; 1127 predState->past_qua_en[1] = ma_pred_init; 1128 predState->past_qua_en[2] = ma_pred_init; 1129 predState->past_qua_en[3] = ma_pred_init; 1130 1131 /* past_qua_en for other modes than MR122 */ 1132 ma_pred_init = mult(5443, ma_pred_init, pOverflow); 1133 /* scale down by factor 20*log10(2) in Q15 */ 1134 predState->past_qua_en_MR122[0] = ma_pred_init; 1135 predState->past_qua_en_MR122[1] = ma_pred_init; 1136 predState->past_qua_en_MR122[2] = ma_pred_init; 1137 predState->past_qua_en_MR122[3] = ma_pred_init; 1138 } /* endif sid_frame */ 1139 1140 /* CN generation */ 1141 /* recompute level adjustment factor Q11 * 1142 * st->log_en_adjust = 0.9*st->log_en_adjust + * 1143 * 0.1*dtx_log_en_adjust[mode]); */ 1144 if (dtx_log_en_adjust[mode] > 1023) 1145 { 1146 temp = MAX_16; 1147 } 1148 else if (dtx_log_en_adjust[mode] < -1024) 1149 { 1150 temp = MIN_16; 1151 } 1152 else 1153 { 1154 temp = mult((Word16)((Word32)dtx_log_en_adjust[mode] << 5), 3277, pOverflow); 1155 } 1156 1157 if (temp < 0) 1158 { 1159 temp = ~((~temp) >> 5); 1160 } 1161 else 1162 { 1163 temp >>= 5; 1164 } 1165 st->log_en_adjust = add(mult(st->log_en_adjust, 29491, pOverflow), temp, pOverflow); 1166 1167 /* Interpolate SID info */ 1168 int_fac = shl(add(1, st->since_last_sid, pOverflow), 10, pOverflow); /* Q10 */ 1169 int_fac = mult(int_fac, st->true_sid_period_inv, pOverflow); /* Q10 * Q15 -> Q10 */ 1170 1171 /* Maximize to 1.0 in Q10 */ 1172 if (int_fac > 1024) 1173 { 1174 int_fac = 16384; 1175 } 1176 else if (int_fac < -2048) 1177 { 1178 int_fac = MIN_16; 1179 } 1180 else 1181 { 1182 int_fac <<= 4; /* Q10 -> Q14 */ 1183 } 1184 1185 L_log_en_int = L_mult(int_fac, st->log_en, pOverflow); /* Q14 * Q11->Q26 */ 1186 for (i = M - 1; i >= 0; i--) 1187 { 1188 lsp_int[i] = mult(int_fac, st->lsp[i], pOverflow);/* Q14 * Q15 -> Q14 */ 1189 } 1190 1191 int_fac = sub(16384, int_fac, pOverflow); /* 1-k in Q14 */ 1192 1193 /* (Q14 * Q11 -> Q26) + Q26 -> Q26 */ 1194 L_log_en_int = L_mac(L_log_en_int, int_fac, st->old_log_en, pOverflow); 1195 for (i = M - 1; i >= 0; i--) 1196 { 1197 /* Q14 + (Q14 * Q15 -> Q14) -> Q14 */ 1198 lsp_int[i] = add(lsp_int[i], mult(int_fac, st->lsp_old[i], pOverflow), pOverflow); 1199 1200 L_temp = ((Word32) lsp_int[i]) << 1; /* Q14 -> Q15 */ 1201 if (L_temp != (Word32)((Word16) L_temp)) 1202 { 1203 *pOverflow = 1; 1204 L_temp = (Word32)((lsp_int[i] > 0) ? MAX_16 : MIN_16); 1205 } 1206 lsp_int[i] = (Word16) L_temp; 1207 } 1208 1209 /* compute the amount of lsf variability */ 1210 lsf_variab_factor = sub(st->log_pg_mean, 2457, pOverflow); /* -0.6 in Q12 */ 1211 /* *0.3 Q12*Q15 -> Q12 */ 1212 lsf_variab_factor = sub(4096, mult(lsf_variab_factor, 9830, pOverflow), pOverflow); 1213 1214 /* limit to values between 0..1 in Q12 */ 1215 if (lsf_variab_factor > 4095) 1216 { 1217 lsf_variab_factor = MAX_16; 1218 } 1219 else if (lsf_variab_factor < 0) 1220 { 1221 lsf_variab_factor = 0; 1222 } 1223 else 1224 { 1225 lsf_variab_factor <<= 3; /* -> Q15 */ 1226 } 1227 1228 /* get index of vector to do variability with */ 1229 lsf_variab_index = pseudonoise(&st->L_pn_seed_rx, 3); 1230 1231 /* convert to lsf */ 1232 Lsp_lsf(lsp_int, lsf_int, M, pOverflow); 1233 1234 /* apply lsf variability */ 1235 Copy(lsf_int, lsf_int_variab, M); 1236 for (i = M - 1; i >= 0; i--) 1237 { 1238 lsf_int_variab[i] = add(lsf_int_variab[i], 1239 mult(lsf_variab_factor, 1240 st->lsf_hist_mean[i+lsf_variab_index*M], pOverflow) 1241 , pOverflow); 1242 } 1243 1244 /* make sure that LSP's are ordered */ 1245 Reorder_lsf(lsf_int, LSF_GAP, M, pOverflow); 1246 Reorder_lsf(lsf_int_variab, LSF_GAP, M, pOverflow); 1247 1248 /* copy lsf to speech decoders lsf state */ 1249 Copy(lsf_int, lsfState->past_lsf_q, M); 1250 1251 /* convert to lsp */ 1252 Lsf_lsp(lsf_int, lsp_int, M, pOverflow); 1253 Lsf_lsp(lsf_int_variab, lsp_int_variab, M, pOverflow); 1254 1255 /* Compute acoeffs Q12 acoeff is used for level * 1256 * normalization and postfilter, acoeff_variab is * 1257 * used for synthesis filter * 1258 * by doing this we make sure that the level * 1259 * in high frequenncies does not jump up and down */ 1260 1261 Lsp_Az(lsp_int, acoeff, pOverflow); 1262 Lsp_Az(lsp_int_variab, acoeff_variab, pOverflow); 1263 1264 /* For use in postfilter */ 1265 Copy(acoeff, &A_t[0], M + 1); 1266 Copy(acoeff, &A_t[M + 1], M + 1); 1267 Copy(acoeff, &A_t[2 *(M + 1)], M + 1); 1268 Copy(acoeff, &A_t[3 *(M + 1)], M + 1); 1269 1270 /* Compute reflection coefficients Q15 */ 1271 A_Refl(&acoeff[1], refl, pOverflow); 1272 1273 /* Compute prediction error in Q15 */ 1274 pred_err = MAX_16; /* 0.99997 in Q15 */ 1275 for (i = 0; i < M; i++) 1276 { 1277 L_temp = (((Word32) refl[i]) * refl[i]) >> 15; 1278 if (L_temp <= 0x00007fffL) 1279 { 1280 temp = MAX_16 - (Word16) L_temp; 1281 } 1282 else 1283 { 1284 *pOverflow = 1; 1285 temp = 0; 1286 } 1287 pred_err = mult(pred_err, temp, pOverflow); 1288 } 1289 1290 /* compute logarithm of prediction gain */ 1291 Log2(L_deposit_l(pred_err), &log_pg_e, &log_pg_m, pOverflow); 1292 1293 /* convert exponent and mantissa to Word16 Q12 */ 1294 log_pg = shl(sub(log_pg_e, 15, pOverflow), 12, pOverflow); /* Q12 */ 1295 log_pg = shr(sub(0, add(log_pg, shr(log_pg_m, 15 - 12, pOverflow), 1296 pOverflow), pOverflow), 1, pOverflow); 1297 st->log_pg_mean = add(mult(29491, st->log_pg_mean, pOverflow), 1298 mult(3277, log_pg, pOverflow), pOverflow); 1299 1300 /* Compute interpolated log energy */ 1301 L_log_en_int = L_shr(L_log_en_int, 10, pOverflow); /* Q26 -> Q16 */ 1302 1303 /* Add 4 in Q16 */ 1304 L_log_en_int = L_add(L_log_en_int, 4 * 65536L, pOverflow); 1305 1306 /* subtract prediction gain */ 1307 L_log_en_int = L_sub(L_log_en_int, L_shl(L_deposit_l(log_pg), 4, pOverflow), pOverflow); 1308 1309 /* adjust level to speech coder mode */ 1310 L_log_en_int = L_add(L_log_en_int, 1311 L_shl(L_deposit_l(st->log_en_adjust), 5, pOverflow), pOverflow); 1312 1313 log_en_int_e = (Word16)(L_log_en_int >> 16); 1314 1315 log_en_int_m = (Word16)(L_shr(L_sub(L_log_en_int, 1316 L_deposit_h(log_en_int_e), pOverflow), 1, pOverflow)); 1317 level = (Word16)(Pow2(log_en_int_e, log_en_int_m, pOverflow)); /* Q4 */ 1318 1319 for (i = 0; i < 4; i++) 1320 { 1321 /* Compute innovation vector */ 1322 build_CN_code(&st->L_pn_seed_rx, ex, pOverflow); 1323 for (j = L_SUBFR - 1; j >= 0; j--) 1324 { 1325 ex[j] = mult(level, ex[j], pOverflow); 1326 } 1327 /* Synthesize */ 1328 Syn_filt(acoeff_variab, ex, &synth[i * L_SUBFR], L_SUBFR, 1329 mem_syn, 1); 1330 1331 } /* next i */ 1332 1333 /* reset codebook averaging variables */ 1334 averState->hangVar = 20; 1335 averState->hangCount = 0; 1336 1337 if (new_state == DTX_MUTE) 1338 { 1339 /* mute comfort noise as it has been quite a long time since 1340 * last SID update was performed */ 1341 1342 tmp_int_length = st->since_last_sid; 1343 1344 if (tmp_int_length > 32) 1345 { 1346 tmp_int_length = 32; 1347 } 1348 else if (tmp_int_length <= 0) 1349 { 1350 /* safety guard against division by zero */ 1351 tmp_int_length = 8; 1352 } 1353 1354 L_temp = ((Word32) tmp_int_length) << 10; 1355 if (L_temp != (Word32)((Word16) L_temp)) 1356 { 1357 *pOverflow = 1; 1358 L_temp = (Word32)((tmp_int_length > 0) ? MAX_16 : MIN_16); 1359 } 1360 temp = (Word16) L_temp; 1361 1362 st->true_sid_period_inv = div_s(1 << 10, temp); 1363 1364 st->since_last_sid = 0; 1365 Copy(st->lsp, st->lsp_old, M); 1366 st->old_log_en = st->log_en; 1367 /* subtract 1/8 in Q11 i.e -6/8 dB */ 1368 st->log_en = sub(st->log_en, 256, pOverflow); 1369 } 1370 1371 /* reset interpolation length timer 1372 * if data has been updated. */ 1373 if ((st->sid_frame != 0) && 1374 ((st->valid_data != 0) || 1375 ((st->valid_data == 0) && (st->dtxHangoverAdded) != 0))) 1376 { 1377 st->since_last_sid = 0; 1378 st->data_updated = 1; 1379 } 1380 1381 return; 1382 } 1383 1384 1385 /****************************************************************************/ 1386 1387 /* 1388 ------------------------------------------------------------------------------ 1389 FUNCTION NAME: dtx_dec_activity_update 1390 ------------------------------------------------------------------------------ 1391 INPUT AND OUTPUT DEFINITIONS 1392 1393 Inputs: 1394 st = pointer to a structure of type dtx_decState 1395 lsf = 1396 frame = 1397 1398 Outputs: 1399 st points to an updated structure of type dtx_decState 1400 1401 Returns: 1402 None. 1403 1404 Global Variables Used: 1405 None. 1406 1407 Local Variables Needed: 1408 None. 1409 1410 ------------------------------------------------------------------------------ 1411 FUNCTION DESCRIPTION 1412 1413 This function updates the DTX parameters. 1414 1415 ------------------------------------------------------------------------------ 1416 REQUIREMENTS 1417 1418 None 1419 1420 ------------------------------------------------------------------------------ 1421 REFERENCES 1422 1423 dtx_dec.c, UMTS GSM AMR speech codec, R99 - Version 3.3.0, December 12, 2001 1424 1425 ------------------------------------------------------------------------------ 1426 PSEUDO-CODE 1427 1428 void dtx_dec_activity_update(dtx_decState *st, 1429 Word16 lsf[], 1430 Word16 frame[]) 1431 { 1432 Word16 i; 1433 1434 Word32 L_frame_en; 1435 Word16 log_en_e, log_en_m, log_en; 1436 1437 // update lsp history 1438 st->lsf_hist_ptr = add(st->lsf_hist_ptr,M); 1439 if (sub(st->lsf_hist_ptr, 80) == 0) 1440 { 1441 st->lsf_hist_ptr = 0; 1442 } 1443 Copy(lsf, &st->lsf_hist[st->lsf_hist_ptr], M); 1444 1445 // compute log energy based on frame energy 1446 L_frame_en = 0; // Q0 1447 for (i=0; i < L_FRAME; i++) 1448 { 1449 L_frame_en = L_mac(L_frame_en, frame[i], frame[i]); 1450 } 1451 Log2(L_frame_en, &log_en_e, &log_en_m); 1452 1453 // convert exponent and mantissa to Word16 Q10 1454 log_en = shl(log_en_e, 10); // Q10 1455 log_en = add(log_en, shr(log_en_m, 15-10)); 1456 1457 // divide with L_FRAME i.e subtract with log2(L_FRAME) = 7.32193 1458 log_en = sub(log_en, 7497+1024); 1459 1460 // insert into log energy buffer, no division by two as * 1461 * log_en in decoder is Q11 1462 st->log_en_hist_ptr = add(st->log_en_hist_ptr, 1); 1463 if (sub(st->log_en_hist_ptr, DTX_HIST_SIZE) == 0) 1464 { 1465 st->log_en_hist_ptr = 0; 1466 } 1467 st->log_en_hist[st->log_en_hist_ptr] = log_en; // Q11 1468 } 1469 1470 ------------------------------------------------------------------------------ 1471 RESOURCES USED [optional] 1472 1473 When the code is written for a specific target processor the 1474 the resources used should be documented below. 1475 1476 HEAP MEMORY USED: x bytes 1477 1478 STACK MEMORY USED: x bytes 1479 1480 CLOCK CYCLES: (cycle count equation for this function) + (variable 1481 used to represent cycle count for each subroutine 1482 called) 1483 where: (cycle count variable) = cycle count for [subroutine 1484 name] 1485 1486 ------------------------------------------------------------------------------ 1487 CAUTION [optional] 1488 [State any special notes, constraints or cautions for users of this function] 1489 1490 ------------------------------------------------------------------------------ 1491 */ 1492 1493 void dtx_dec_activity_update(dtx_decState *st, 1494 Word16 lsf[], 1495 Word16 frame[], 1496 Flag *pOverflow) 1497 { 1498 Word16 i; 1499 1500 Word32 L_frame_en; 1501 Word32 L_temp; 1502 Word16 log_en_e; 1503 Word16 log_en_m; 1504 Word16 log_en; 1505 1506 /* update lsp history */ 1507 st->lsf_hist_ptr += M; 1508 1509 if (st->lsf_hist_ptr == 80) 1510 { 1511 st->lsf_hist_ptr = 0; 1512 } 1513 Copy(lsf, &st->lsf_hist[st->lsf_hist_ptr], M); 1514 1515 /* compute log energy based on frame energy */ 1516 L_frame_en = 0; /* Q0 */ 1517 for (i = L_FRAME - 1; i >= 0; i--) 1518 { 1519 L_temp = ((Word32) frame[i]) * frame[i]; 1520 if (L_temp != (Word32) 0x40000000L) 1521 { 1522 L_temp = L_temp << 1; 1523 } 1524 else 1525 { 1526 L_temp = MAX_32; 1527 } 1528 L_frame_en = L_add(L_frame_en, L_temp, pOverflow); 1529 } 1530 Log2(L_frame_en, &log_en_e, &log_en_m, pOverflow); 1531 1532 /* convert exponent and mantissa to Word16 Q10 */ 1533 L_temp = ((Word32) log_en_e) << 10; 1534 1535 if (L_temp != (Word32)((Word16) L_temp)) 1536 { 1537 *pOverflow = 1; 1538 L_temp = (Word32)((log_en_e > 0) ? MAX_16 : MIN_16); 1539 } 1540 log_en_e = (Word16) L_temp; 1541 1542 if (log_en_m < 0) 1543 { 1544 log_en_m = ~((~log_en_m) >> 5); 1545 } 1546 else 1547 { 1548 log_en_m >>= 5; 1549 } 1550 log_en = add(log_en_e, log_en_m, pOverflow); 1551 1552 /* divide with L_FRAME i.e subtract with log2(L_FRAME) = 7.32193 */ 1553 log_en = sub(log_en, 7497 + 1024, pOverflow); 1554 1555 /* insert into log energy buffer, no division by two as * 1556 * log_en in decoder is Q11 */ 1557 st->log_en_hist_ptr += 1; 1558 1559 if (st->log_en_hist_ptr == DTX_HIST_SIZE) 1560 { 1561 st->log_en_hist_ptr = 0; 1562 } 1563 st->log_en_hist[st->log_en_hist_ptr] = log_en; /* Q11 */ 1564 1565 return; 1566 } 1567 1568 /****************************************************************************/ 1569 1570 /* 1571 ------------------------------------------------------------------------------ 1572 FUNCTION NAME: rx_dtx_handler 1573 ------------------------------------------------------------------------------ 1574 INPUT AND OUTPUT DEFINITIONS 1575 1576 Inputs: 1577 st = pointer to a structure of type dtx_decState 1578 frame_type = RX frame type 1579 1580 Returns: 1581 newState = variable of type DTXStateType 1582 1583 Outputs: 1584 st points to an updated structure of type dtx_decState 1585 1586 Global Variables Used: 1587 None. 1588 1589 Local Variables Needed: 1590 None. 1591 1592 ------------------------------------------------------------------------------ 1593 FUNCTION DESCRIPTION 1594 1595 This function determines the new state of the decoder based on the frame_type 1596 and sets up the decoder parameters according to newState. 1597 1598 Table of new SPD synthesis states 1599 1600 | previous SPD_synthesis_state 1601 Incoming | 1602 frame_type | SPEECH | DTX | DTX_MUTE 1603 --------------------------------------------------------------- 1604 RX_SPEECH_GOOD , | | | 1605 RX_SPEECH_PR_DEGRADED | SPEECH | SPEECH | SPEECH 1606 ---------------------------------------------------------------- 1607 RX_SPEECH_PR_BAD, | | | 1608 RX_SPEECH_BAD, | SPEECH | DTX | DTX_MUTE 1609 ---------------------------------------------------------------- 1610 RX_SID_FIRST, | DTX | DTX/(DTX_MUTE)| DTX_MUTE 1611 ---------------------------------------------------------------- 1612 RX_SID_UPDATE, | DTX | DTX | DTX 1613 ---------------------------------------------------------------- 1614 RX_SID_BAD, | DTX | DTX/(DTX_MUTE)| DTX_MUTE 1615 ---------------------------------------------------------------- 1616 RX_NO_DATA | SPEECH | DTX/(DTX_MUTE)| DTX_MUTE 1617 |(class2 garb.)| | 1618 ---------------------------------------------------------------- 1619 RX_ONSET | SPEECH | DTX/(DTX_MUTE)| DTX_MUTE 1620 |(class2 garb.)| | 1621 ---------------------------------------------------------------- 1622 1623 ------------------------------------------------------------------------------ 1624 REQUIREMENTS 1625 1626 None 1627 1628 ------------------------------------------------------------------------------ 1629 REFERENCES 1630 1631 dtx_dec.c, UMTS GSM AMR speech codec, R99 - Version 3.3.0, December 12, 2001 1632 1633 ------------------------------------------------------------------------------ 1634 PSEUDO-CODE 1635 1636 enum DTXStateType rx_dtx_handler( 1637 dtx_decState *st, // i/o : State struct 1638 enum RXFrameType frame_type // i : Frame type 1639 ) 1640 { 1641 enum DTXStateType newState; 1642 enum DTXStateType encState; 1643 1644 // DTX if SID frame or previously in DTX{_MUTE} and (NO_RX OR BAD_SPEECH) 1645 if ((sub(frame_type, RX_SID_FIRST) == 0) || 1646 (sub(frame_type, RX_SID_UPDATE) == 0) || 1647 (sub(frame_type, RX_SID_BAD) == 0) || 1648 (((sub(st->dtxGlobalState, DTX) == 0) || 1649 (sub(st->dtxGlobalState, DTX_MUTE) == 0)) && 1650 ((sub(frame_type, RX_NO_DATA) == 0) || 1651 (sub(frame_type, RX_SPEECH_BAD) == 0) || 1652 (sub(frame_type, RX_ONSET) == 0)))) 1653 { 1654 newState = DTX; 1655 1656 // stay in mute for these input types 1657 if ((sub(st->dtxGlobalState, DTX_MUTE) == 0) && 1658 ((sub(frame_type, RX_SID_BAD) == 0) || 1659 (sub(frame_type, RX_SID_FIRST) == 0) || 1660 (sub(frame_type, RX_ONSET) == 0) || 1661 (sub(frame_type, RX_NO_DATA) == 0))) 1662 { 1663 newState = DTX_MUTE; 1664 } 1665 1666 // evaluate if noise parameters are too old 1667 // since_last_sid is reset when CN parameters have been updated 1668 st->since_last_sid = add(st->since_last_sid, 1); 1669 1670 // no update of sid parameters in DTX for a long while 1671 // Due to the delayed update of st->since_last_sid counter 1672 // SID_UPDATE frames need to be handled separately to avoid 1673 // entering DTX_MUTE for late SID_UPDATE frames 1674 if((sub(frame_type, RX_SID_UPDATE) != 0) && 1675 (sub(st->since_last_sid, DTX_MAX_EMPTY_THRESH) > 0)) 1676 { 1677 newState = DTX_MUTE; 1678 } 1679 } 1680 else 1681 { 1682 newState = SPEECH; 1683 st->since_last_sid = 0; 1684 } 1685 1686 // reset the decAnaElapsed Counter when receiving CNI data the first 1687 // time, to robustify counter missmatch after handover 1688 // this might delay the bwd CNI analysis in the new decoder slightly. 1689 1690 if ((st->data_updated == 0) && 1691 (sub(frame_type, RX_SID_UPDATE) == 0)) 1692 { 1693 st->decAnaElapsedCount = 0; 1694 } 1695 1696 // update the SPE-SPD DTX hangover synchronization 1697 // to know when SPE has added dtx hangover 1698 st->decAnaElapsedCount = add(st->decAnaElapsedCount, 1); 1699 st->dtxHangoverAdded = 0; 1700 1701 if ((sub(frame_type, RX_SID_FIRST) == 0) || 1702 (sub(frame_type, RX_SID_UPDATE) == 0) || 1703 (sub(frame_type, RX_SID_BAD) == 0) || 1704 (sub(frame_type, RX_ONSET) == 0) || 1705 (sub(frame_type, RX_NO_DATA) == 0)) 1706 { 1707 encState = DTX; 1708 1709 // In frame errors simulations RX_NO_DATA may occasionally mean that 1710 // a speech packet was probably sent by the encoder, 1711 // the assumed _encoder_ state should be SPEECH in such cases. 1712 if((sub(frame_type, RX_NO_DATA) == 0) && 1713 (sub(newState, SPEECH) == 0)) 1714 { 1715 encState = SPEECH; 1716 } 1717 1718 // Note on RX_ONSET operation differing from RX_NO_DATA operation: 1719 // If a RX_ONSET is received in the decoder (by "accident") 1720 // it is still most likely that the encoder state 1721 // for the "ONSET frame" was DTX. 1722 1723 } 1724 else 1725 { 1726 encState = SPEECH; 1727 } 1728 1729 if (sub(encState, SPEECH) == 0) 1730 { 1731 st->dtxHangoverCount = DTX_HANG_CONST; 1732 } 1733 else 1734 { 1735 if (sub(st->decAnaElapsedCount, DTX_ELAPSED_FRAMES_THRESH) > 0) 1736 { 1737 st->dtxHangoverAdded = 1; 1738 st->decAnaElapsedCount = 0; 1739 st->dtxHangoverCount = 0; 1740 } 1741 else if (st->dtxHangoverCount == 0) 1742 { 1743 st->decAnaElapsedCount = 0; 1744 } 1745 else 1746 { 1747 st->dtxHangoverCount = sub(st->dtxHangoverCount, 1); 1748 } 1749 } 1750 1751 if (sub(newState, SPEECH) != 0) 1752 { 1753 // DTX or DTX_MUTE 1754 // CN data is not in a first SID, first SIDs are marked as SID_BAD 1755 // but will do backwards analysis if a hangover period has been added 1756 // according to the state machine above 1757 1758 st->sid_frame = 0; 1759 st->valid_data = 0; 1760 1761 if (sub(frame_type, RX_SID_FIRST) == 0) 1762 { 1763 st->sid_frame = 1; 1764 } 1765 else if (sub(frame_type, RX_SID_UPDATE) == 0) 1766 { 1767 st->sid_frame = 1; 1768 st->valid_data = 1; 1769 } 1770 else if (sub(frame_type, RX_SID_BAD) == 0) 1771 { 1772 st->sid_frame = 1; 1773 st->dtxHangoverAdded = 0; // use old data 1774 } 1775 } 1776 1777 return newState; 1778 // newState is used by both SPEECH AND DTX synthesis routines 1779 } 1780 1781 ------------------------------------------------------------------------------ 1782 RESOURCES USED [optional] 1783 1784 When the code is written for a specific target processor the 1785 the resources used should be documented below. 1786 1787 HEAP MEMORY USED: x bytes 1788 1789 STACK MEMORY USED: x bytes 1790 1791 CLOCK CYCLES: (cycle count equation for this function) + (variable 1792 used to represent cycle count for each subroutine 1793 called) 1794 where: (cycle count variable) = cycle count for [subroutine 1795 name] 1796 1797 ------------------------------------------------------------------------------ 1798 CAUTION [optional] 1799 [State any special notes, constraints or cautions for users of this function] 1800 1801 ------------------------------------------------------------------------------ 1802 */ 1803 1804 enum DTXStateType rx_dtx_handler( 1805 dtx_decState *st, /* i/o : State struct */ 1806 enum RXFrameType frame_type,/* i : Frame type */ 1807 Flag *pOverflow) 1808 { 1809 enum DTXStateType newState; 1810 enum DTXStateType encState; 1811 1812 1813 /* DTX if SID frame or previously in DTX{_MUTE} and (NO_RX OR BAD_SPEECH) */ 1814 1815 if ((frame_type == RX_SID_FIRST) || 1816 (frame_type == RX_SID_UPDATE) || 1817 (frame_type == RX_SID_BAD) || 1818 (((st->dtxGlobalState == DTX) || (st->dtxGlobalState == DTX_MUTE)) && 1819 ((frame_type == RX_NO_DATA) || (frame_type == RX_SPEECH_BAD) || 1820 (frame_type == RX_ONSET)))) 1821 { 1822 newState = DTX; 1823 1824 /* stay in mute for these input types */ 1825 1826 if ((st->dtxGlobalState == DTX_MUTE) && 1827 ((frame_type == RX_SID_BAD) || 1828 (frame_type == RX_SID_FIRST) || 1829 (frame_type == RX_ONSET) || 1830 (frame_type == RX_NO_DATA))) 1831 { 1832 newState = DTX_MUTE; 1833 } 1834 1835 /* evaluate if noise parameters are too old */ 1836 /* since_last_sid is reset when CN parameters have been updated */ 1837 st->since_last_sid = add(st->since_last_sid, 1, pOverflow); 1838 1839 /* no update of sid parameters in DTX for a long while */ 1840 /* Due to the delayed update of st->since_last_sid counter */ 1841 /* SID_UPDATE frames need to be handled separately to avoid */ 1842 /* entering DTX_MUTE for late SID_UPDATE frames */ 1843 if ((frame_type != RX_SID_UPDATE) && 1844 (st->since_last_sid > DTX_MAX_EMPTY_THRESH)) 1845 { 1846 newState = DTX_MUTE; 1847 } 1848 } 1849 else 1850 { 1851 newState = SPEECH; 1852 st->since_last_sid = 0; 1853 } 1854 1855 /* 1856 reset the decAnaElapsed Counter when receiving CNI data the first 1857 time, to robustify counter missmatch after handover 1858 this might delay the bwd CNI analysis in the new decoder slightly. 1859 */ 1860 1861 if ((st->data_updated == 0) && 1862 (frame_type == RX_SID_UPDATE)) 1863 { 1864 st->decAnaElapsedCount = 0; 1865 } 1866 1867 /* update the SPE-SPD DTX hangover synchronization */ 1868 /* to know when SPE has added dtx hangover */ 1869 st->decAnaElapsedCount = add(st->decAnaElapsedCount, 1, pOverflow); 1870 st->dtxHangoverAdded = 0; 1871 1872 if ((frame_type == RX_SID_FIRST) || 1873 (frame_type == RX_SID_UPDATE) || 1874 (frame_type == RX_SID_BAD) || 1875 (frame_type == RX_ONSET) || 1876 (frame_type == RX_NO_DATA)) 1877 { 1878 encState = DTX; 1879 1880 /* 1881 In frame errors simulations RX_NO_DATA may occasionally mean that 1882 a speech packet was probably sent by the encoder, 1883 the assumed _encoder_ state should be SPEECH in such cases. 1884 */ 1885 if ((frame_type == RX_NO_DATA) && 1886 (newState == SPEECH)) 1887 { 1888 encState = SPEECH; 1889 } 1890 1891 /* 1892 Note on RX_ONSET operation differing from RX_NO_DATA operation: 1893 If a RX_ONSET is received in the decoder (by "accident") 1894 it is still most likely that the encoder state 1895 for the "ONSET frame" was DTX. 1896 */ 1897 } 1898 else 1899 { 1900 encState = SPEECH; 1901 } 1902 1903 1904 if (encState == SPEECH) 1905 { 1906 st->dtxHangoverCount = DTX_HANG_CONST; 1907 } 1908 else 1909 { 1910 1911 if (st->decAnaElapsedCount > DTX_ELAPSED_FRAMES_THRESH) 1912 { 1913 st->dtxHangoverAdded = 1; 1914 st->decAnaElapsedCount = 0; 1915 st->dtxHangoverCount = 0; 1916 } 1917 else if (st->dtxHangoverCount == 0) 1918 { 1919 st->decAnaElapsedCount = 0; 1920 } 1921 else 1922 { 1923 st->dtxHangoverCount -= 1; 1924 } 1925 } 1926 1927 if (newState != SPEECH) 1928 { 1929 /* DTX or DTX_MUTE 1930 * CN data is not in a first SID, first SIDs are marked as SID_BAD 1931 * but will do backwards analysis if a hangover period has been added 1932 * according to the state machine above 1933 */ 1934 1935 st->sid_frame = 0; 1936 st->valid_data = 0; 1937 1938 if (frame_type == RX_SID_FIRST) 1939 { 1940 st->sid_frame = 1; 1941 } 1942 else if (frame_type == RX_SID_UPDATE) 1943 { 1944 st->sid_frame = 1; 1945 st->valid_data = 1; 1946 } 1947 else if (frame_type == RX_SID_BAD) 1948 { 1949 st->sid_frame = 1; 1950 st->dtxHangoverAdded = 0; /* use old data */ 1951 } 1952 } 1953 1954 /* newState is used by both SPEECH AND DTX synthesis routines */ 1955 return(newState); 1956 } 1957
