From f02dfce6e6c34b3d8a7b8a0e784b506178e331fa Mon Sep 17 00:00:00 2001 From: "erdgeist@erdgeist.org" Date: Thu, 4 Jul 2019 23:26:09 +0200 Subject: stripdown of version 0.9 --- newamp2.c | 570 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 570 insertions(+) create mode 100644 newamp2.c (limited to 'newamp2.c') diff --git a/newamp2.c b/newamp2.c new file mode 100644 index 0000000..6550557 --- /dev/null +++ b/newamp2.c @@ -0,0 +1,570 @@ +/*---------------------------------------------------------------------------*\ + + FILE........: newamp2.c + AUTHOR......: Thomas Kurin and Stefan Erhardt + INSTITUTE...: Institute for Electronics Engineering, University of Erlangen-Nuremberg + DATE CREATED: July 2018 + BASED ON....: "newamp1" by David Rowe + + Quantisation functions for the sinusoidal coder, using "newamp1" + algorithm that resamples variable rate L [Am} to a fixed rate K then + VQs. + +\*---------------------------------------------------------------------------*/ + +/* + Copyright David Rowe 2017 + + All rights reserved. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU Lesser General Public License version 2.1, as + published by the Free Software Foundation. This program is + distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public + License for more details. + + You should have received a copy of the GNU Lesser General Public License + along with this program; if not, see . + +*/ + +#include +#include +#include +#include +#include + +#include "defines.h" +#include "phase.h" +#include "quantise.h" +#include "mbest.h" +#include "newamp1.h" +#include "newamp2.h" + +/*---------------------------------------------------------------------------*\ + + FUNCTION....: n2_mel_sample_freqs_kHz() + AUTHOR......: Thomas Kurin and Stefan Erhardt + INSTITUTE...: Institute for Electronics Engineering, University of Erlangen-Nuremberg + DATE CREATED: July 2018 + + Outputs fixed frequencies for the K-Vectors to be able to work with both 8k and 16k mode. + +\*---------------------------------------------------------------------------*/ + +void n2_mel_sample_freqs_kHz(float rate_K_sample_freqs_kHz[], int K) +{ + float freq[] = {0.199816, 0.252849, 0.309008, 0.368476, 0.431449, 0.498134, 0.568749, 0.643526, 0.722710, 0.806561, 0.895354, 0.989380, + 1.088948, 1.194384, 1.306034, 1.424264, 1.549463, 1.682041, 1.822432, 1.971098, 2.128525, 2.295232, 2.471763, 2.658699, + 2.856652, 3.066272, 3.288246, 3.523303, 3.772214, 4.035795, 4.314912, 4.610478, 4.923465, 5.254899, 5.605865, 5.977518, + 6.371075, 6.787827, 7.229141, 7.696465}; + int k; + //printf("\n\n"); + for (k=0; kL; m++) { + AmdB[m] = 20.0*log10(model->A[m]+1E-16); + if (AmdB[m] > AmdB_peak) { + AmdB_peak = AmdB[m]; + } + rate_L_sample_freqs_kHz[m] = m*model->Wo*(c2const->Fs/2000.0)/M_PI; + //printf("m: %d AmdB: %f AmdB_peak: %f sf: %f\n", m, AmdB[m], AmdB_peak, rate_L_sample_freqs_kHz[m]); + } + + /* clip between peak and peak -50dB, to reduce dynamic range */ + + for(m=1; m<=model->L; m++) { + if (AmdB[m] < (AmdB_peak-50.0)) { + AmdB[m] = AmdB_peak-50.0; + } + } + + interp_para(rate_K_vec, &rate_L_sample_freqs_kHz[1], &AmdB[1], model->L, rate_K_sample_freqs_kHz, K); +} + + +/*---------------------------------------------------------------------------*\ + + FUNCTION....: n2_rate_K_mbest_encode + AUTHOR......: Thomas Kurin and Stefan Erhardt + INSTITUTE...: Institute for Electronics Engineering, University of Erlangen-Nuremberg + DATE CREATED: July 2018 + + One stage rate K newamp2 VQ quantiser using mbest search. + +\*---------------------------------------------------------------------------*/ + +void n2_rate_K_mbest_encode(int *indexes, float *x, float *xq, int ndim) +{ + int i, n1; + const float *codebook1 = newamp2vq_cb[0].cb; + struct MBEST *mbest_stage1; + float w[ndim]; + int index[1]; + + /* codebook is compiled for a fixed K */ + + //assert(ndim == newamp2vq_cb[0].k); + + /* equal weights, could be argued mel freq axis gives freq dep weighting */ + + for(i=0; ilist[0].index[0]; + + mbest_destroy(mbest_stage1); + + //indexes[1]: legacy from newamp1 + indexes[0] = n1; indexes[1] = n1; + +} + + +/*---------------------------------------------------------------------------*\ + + FUNCTION....: n2_resample_rate_L + AUTHOR......: Thomas Kurin and Stefan Erhardt + INSTITUTE...: Institute for Electronics Engineering, University of Erlangen-Nuremberg + DATE CREATED: July 2018 + + Decoder side conversion of rate K vector back to rate L. + Plosives are set to zero for the first 2 of 4 frames. + +\*---------------------------------------------------------------------------*/ + +void n2_resample_rate_L(C2CONST *c2const, MODEL *model, float rate_K_vec[], float rate_K_sample_freqs_kHz[], int K,int plosive_flag) +{ + float rate_K_vec_term[K+2], rate_K_sample_freqs_kHz_term[K+2]; + float AmdB[MAX_AMP+1], rate_L_sample_freqs_kHz[MAX_AMP+1]; + int m,k; + + /* terminate either end of the rate K vecs with 0dB points */ + + rate_K_vec_term[0] = rate_K_vec_term[K+1] = 0.0; + rate_K_sample_freqs_kHz_term[0] = 0.0; + rate_K_sample_freqs_kHz_term[K+1] = 4.0; + + for(k=0; kL; m++) { + rate_L_sample_freqs_kHz[m] = m*model->Wo*(c2const->Fs/2000.0)/M_PI; + } + + interp_para(&AmdB[1], rate_K_sample_freqs_kHz_term, rate_K_vec_term, K+2, &rate_L_sample_freqs_kHz[1], model->L); + for(m=1; m<=model->L; m++) { + if(plosive_flag==0){ + model->A[m] = pow(10.0, AmdB[m]/20.0); + }else{ + model->A[m] = 0.1; + } + // printf("m: %d f: %f AdB: %f A: %f\n", m, rate_L_sample_freqs_kHz[m], AmdB[m], model->A[m]); + } +} + +/*---------------------------------------------------------------------------*\ + + FUNCTION....: n2_post_filter_newamp2 + AUTHOR......: Thomas Kurin and Stefan Erhardt + INSTITUTE...: Institute for Electronics Engineering, University of Erlangen-Nuremberg + DATE CREATED: July 2018 + + Postfilter for the pseudo wideband mode. Still has to be adapted! + +\*---------------------------------------------------------------------------*/ + +void n2_post_filter_newamp2(float vec[], float sample_freq_kHz[], int K, float pf_gain) +{ + int k; + + /* + vec is rate K vector describing spectrum of current frame lets + pre-emp before applying PF. 20dB/dec over 300Hz. Postfilter + affects energy of frame so we measure energy before and after + and normalise. Plenty of room for experiment here as well. + */ + + float pre[K]; + float e_before = 0.0; + float e_after = 0.0; + for(k=0; kvoiced) { + int index = encode_log_Wo(c2const, model->Wo, 6); + if (index == 0) { + index = 1; + } + if (index == 63) { + index = 62; + } + indexes[3] = index; + } + else { + indexes[3] = 0; + } + if(plosive != 0){ + indexes[3] = 63; + } + } + + +/*---------------------------------------------------------------------------*\ + + FUNCTION....: newamp2_indexes_to_rate_K_vec + AUTHOR......: Thomas Kurin and Stefan Erhardt + INSTITUTE...: Institute for Electronics Engineering, University of Erlangen-Nuremberg + DATE CREATED: July 2018 + + newamp2 decoder for amplitudes {Am}. Given the rate K VQ and energy + indexes, outputs rate K vector. Equal to newamp1 but using only one stage VQ. + +\*---------------------------------------------------------------------------*/ + +void newamp2_indexes_to_rate_K_vec(float rate_K_vec_[], + float rate_K_vec_no_mean_[], + float rate_K_sample_freqs_kHz[], + int K, + float *mean_, + int indexes[], + float pf_gain) +{ + int k; + const float *codebook1 = newamp2vq_cb[0].cb; + int n1 = indexes[0]; + + for(k=0; kmean2) + +\*---------------------------------------------------------------------------*/ + +void newamp2_16k_indexes_to_rate_K_vec(float rate_K_vec_[], + float rate_K_vec_no_mean_[], + float rate_K_sample_freqs_kHz[], + int K, + float *mean_, + int indexes[], + float pf_gain) +{ + int k; + const float *codebook1 = newamp2vq_cb[0].cb; + float mean2 = 0; + int n1 = indexes[0]; + + for(k=0; k50){ + mean2 = 50; + } + + for(k=0; k0 && indexes[3]<63) { + Wo_right = decode_log_Wo(c2const, indexes[3], 6); + voicing_right = 1; + } + //Unvoiced + else if(indexes[3] == 0){ + Wo_right = 2.0*M_PI/100.0; + voicing_right = 0; + } + //indexes[3]=63 (= Plosive) and unvoiced + else { + Wo_right = 2.0*M_PI/100.0; + voicing_right = 0; + plosive_flag = 1; + } + + /* interpolate 25Hz rate K vec back to 100Hz */ + + float *left_vec = prev_rate_K_vec_; + float *right_vec = rate_K_vec_; + newamp2_interpolate(interpolated_surface_, left_vec, right_vec, K,plosive_flag); + + /* interpolate 25Hz v and Wo back to 100Hz */ + + float aWo_[M]; + int avoicing_[M], aL_[M], i; + + interp_Wo_v(aWo_, aL_, avoicing_, *Wo_left, Wo_right, *voicing_left, voicing_right); + + /* back to rate L amplitudes, synthesis phase for each frame */ + + for(i=0; i0){ + //First two frames are set to zero + if (i<2){ + n2_resample_rate_L(c2const, &model_[i], &interpolated_surface_[K*i], rate_K_sample_freqs_kHz, K,1); + } + else{ + n2_resample_rate_L(c2const, &model_[i], &interpolated_surface_[K*i], rate_K_sample_freqs_kHz, K,0); + } + } + //No Plosive, standard resample + else{ + n2_resample_rate_L(c2const, &model_[i], &interpolated_surface_[K*i], rate_K_sample_freqs_kHz, K,0); + } + determine_phase(c2const, &H[(MAX_AMP+1)*i], &model_[i], NEWAMP2_PHASE_NFFT, fwd_cfg, inv_cfg); + } + + /* update memories for next time */ + + for(k=0; k