diff --git a/CMakeLists.txt b/CMakeLists.txt index 86d55992..d296efe2 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -1146,14 +1146,6 @@ endif(NOT APPLE) COMMAND sh -c "./c2enc 700C ${CMAKE_CURRENT_SOURCE_DIR}/raw/hts1a.raw - | ./c2dec 700C - - | sox -t .s16 -r 8000 - hts1a_700C.wav" WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/src ) - add_test(NAME test_codec2_mode_450 - COMMAND sh -c "./c2enc 450 ${CMAKE_CURRENT_SOURCE_DIR}/raw/hts1a.raw - | ./c2dec 450 - - | sox -t .s16 -r 8000 - hts1a_450.wav" - WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/src - ) - add_test(NAME test_codec2_mode_450PWB - COMMAND sh -c "./c2enc 450PWB ${CMAKE_CURRENT_SOURCE_DIR}/raw/hts1a.raw - | ./c2dec 450PWB - - | sox -t .s16 -r 16000 - hts1a_450PWB.wav" - WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/src - ) add_test(NAME test_vq_mbest COMMAND sh -c "./tvq_mbest; \ diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt index 25081a0e..fb10d287 100644 --- a/src/CMakeLists.txt +++ b/src/CMakeLists.txt @@ -169,7 +169,6 @@ set(CODEC2_SRCS lsp.c mbest.c newamp1.c - newamp2.c ofdm.c ofdm_mode.c phase.c diff --git a/src/c2dec.c b/src/c2dec.c index 5cd5383e..b852078e 100644 --- a/src/c2dec.c +++ b/src/c2dec.c @@ -139,14 +139,10 @@ int main(int argc, char *argv[]) { mode = CODEC2_MODE_1200; else if (strcmp(argv[1], "700C") == 0) mode = CODEC2_MODE_700C; - else if (strcmp(argv[1], "450") == 0) - mode = CODEC2_MODE_450; - else if (strcmp(argv[1], "450PWB") == 0) - mode = CODEC2_MODE_450PWB; else { fprintf(stderr, "Error in mode: %s. Must be 3200, 2400, 1600, 1400, 1300, 1200, " - "700C, 450, or 450PWB\n", + "700C\n", argv[1]); exit(1); } @@ -396,7 +392,7 @@ void print_help(const struct option *long_options, int num_opts, char *argv[]) { char *option_parameters; fprintf(stderr, "\nc2dec - Codec 2 decoder and bit error simulation program\n" - "usage: %s 3200|2400|1600|1400|1300|1200|700C|450|450PWB InputFile " + "usage: %s 3200|2400|1600|1400|1300|1200|700C InputFile " "OutputRawFile [OPTIONS]\n\n" "Options:\n", argv[0]); diff --git a/src/c2enc.c b/src/c2enc.c index fe37acd1..5511d9dd 100644 --- a/src/c2enc.c +++ b/src/c2enc.c @@ -51,7 +51,7 @@ int main(int argc, char *argv[]) { if (argc < 4) { printf( - "usage: c2enc 3200|2400|1600|1400|1300|1200|700C|450|450PWB " + "usage: c2enc 3200|2400|1600|1400|1300|1200|700C " "InputRawspeechFile OutputBitFile [--natural] [--softdec] " "[--bitperchar] [--mlfeat f32File modelFile] [--loadcb stageNum " "Filename] [--var] [--eq]\n"); @@ -76,14 +76,10 @@ int main(int argc, char *argv[]) { mode = CODEC2_MODE_1200; else if (strcmp(argv[1], "700C") == 0) mode = CODEC2_MODE_700C; - else if (strcmp(argv[1], "450") == 0) - mode = CODEC2_MODE_450; - else if (strcmp(argv[1], "450PWB") == 0) - mode = CODEC2_MODE_450; else { fprintf(stderr, "Error in mode: %s. Must be 3200, 2400, 1600, 1400, 1300, 1200, " - "700C, 450, 450PWB or WB\n", + "700C\n", argv[1]); exit(1); } diff --git a/src/codec2.c b/src/codec2.c index 264141cd..52602e38 100644 --- a/src/codec2.c +++ b/src/codec2.c @@ -83,11 +83,6 @@ void codec2_decode_1200(struct CODEC2 *c2, short speech[], void codec2_encode_700c(struct CODEC2 *c2, unsigned char *bits, short speech[]); void codec2_decode_700c(struct CODEC2 *c2, short speech[], const unsigned char *bits); -void codec2_encode_450(struct CODEC2 *c2, unsigned char *bits, short speech[]); -void codec2_decode_450(struct CODEC2 *c2, short speech[], - const unsigned char *bits); -void codec2_decode_450pwb(struct CODEC2 *c2, short speech[], - const unsigned char *bits); static void ear_protection(float in_out[], int n); /*---------------------------------------------------------------------------*\ @@ -110,7 +105,6 @@ static void ear_protection(float in_out[], int n); \*---------------------------------------------------------------------------*/ -// Don't create CODEC2_MODE_450PWB for Encoding as it has undefined behavior ! struct CODEC2 *codec2_create(int mode) { struct CODEC2 *c2; int i, l; @@ -125,9 +119,7 @@ struct CODEC2 *codec2_create(int mode) { CODEC2_MODE_ACTIVE(CODEC2_MODE_1400, mode) || CODEC2_MODE_ACTIVE(CODEC2_MODE_1300, mode) || CODEC2_MODE_ACTIVE(CODEC2_MODE_1200, mode) || - CODEC2_MODE_ACTIVE(CODEC2_MODE_700C, mode) || - CODEC2_MODE_ACTIVE(CODEC2_MODE_450, mode) || - CODEC2_MODE_ACTIVE(CODEC2_MODE_450PWB, mode))) { + CODEC2_MODE_ACTIVE(CODEC2_MODE_700C, mode))) { return NULL; } @@ -138,11 +130,7 @@ struct CODEC2 *codec2_create(int mode) { /* store constants in a few places for convenience */ - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450PWB, mode) == 0) { - c2->c2const = c2const_create(8000, N_S); - } else { - c2->c2const = c2const_create(16000, N_S); - } + c2->c2const = c2const_create(8000, N_S); c2->Fs = c2->c2const.Fs; int n_samp = c2->n_samp = c2->c2const.n_samp; int m_pitch = c2->m_pitch = c2->c2const.m_pitch; @@ -231,40 +219,10 @@ struct CODEC2 *codec2_create(int mode) { c2->eq_en = false; c2->Wo_left = 0.0; c2->voicing_left = 0; - ; c2->phase_fft_fwd_cfg = codec2_fft_alloc(NEWAMP1_PHASE_NFFT, 0, NULL, NULL); c2->phase_fft_inv_cfg = codec2_fft_alloc(NEWAMP1_PHASE_NFFT, 1, NULL, NULL); } - /* newamp2 initialisation */ - - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450, c2->mode)) { - n2_mel_sample_freqs_kHz(c2->n2_rate_K_sample_freqs_kHz, NEWAMP2_K); - int k; - for (k = 0; k < NEWAMP2_K; k++) { - c2->n2_prev_rate_K_vec_[k] = 0.0; - } - c2->Wo_left = 0.0; - c2->voicing_left = 0; - ; - c2->phase_fft_fwd_cfg = codec2_fft_alloc(NEWAMP2_PHASE_NFFT, 0, NULL, NULL); - c2->phase_fft_inv_cfg = codec2_fft_alloc(NEWAMP2_PHASE_NFFT, 1, NULL, NULL); - } - /* newamp2 PWB initialisation */ - - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450PWB, c2->mode)) { - n2_mel_sample_freqs_kHz(c2->n2_pwb_rate_K_sample_freqs_kHz, NEWAMP2_16K_K); - int k; - for (k = 0; k < NEWAMP2_16K_K; k++) { - c2->n2_pwb_prev_rate_K_vec_[k] = 0.0; - } - c2->Wo_left = 0.0; - c2->voicing_left = 0; - ; - c2->phase_fft_fwd_cfg = codec2_fft_alloc(NEWAMP2_PHASE_NFFT, 0, NULL, NULL); - c2->phase_fft_inv_cfg = codec2_fft_alloc(NEWAMP2_PHASE_NFFT, 1, NULL, NULL); - } - c2->fmlfeat = NULL; c2->fmlmodel = NULL; @@ -310,17 +268,6 @@ struct CODEC2 *codec2_create(int mode) { c2->decode = codec2_decode_700c; } - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450, c2->mode)) { - c2->encode = codec2_encode_450; - c2->decode = codec2_decode_450; - } - - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450PWB, c2->mode)) { - // Encode PWB doesn't make sense - c2->encode = codec2_encode_450; - c2->decode = codec2_decode_450pwb; - } - return c2; } @@ -345,14 +292,6 @@ void codec2_destroy(struct CODEC2 *c2) { codec2_fft_free(c2->phase_fft_fwd_cfg); codec2_fft_free(c2->phase_fft_inv_cfg); } - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450, c2->mode)) { - codec2_fft_free(c2->phase_fft_fwd_cfg); - codec2_fft_free(c2->phase_fft_inv_cfg); - } - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450PWB, c2->mode)) { - codec2_fft_free(c2->phase_fft_fwd_cfg); - codec2_fft_free(c2->phase_fft_inv_cfg); - } FREE(c2->Pn); FREE(c2->Sn); FREE(c2->w); @@ -378,8 +317,6 @@ int codec2_bits_per_frame(struct CODEC2 *c2) { if (CODEC2_MODE_ACTIVE(CODEC2_MODE_1300, c2->mode)) return 52; if (CODEC2_MODE_ACTIVE(CODEC2_MODE_1200, c2->mode)) return 48; if (CODEC2_MODE_ACTIVE(CODEC2_MODE_700C, c2->mode)) return 28; - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450, c2->mode)) return 18; - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450PWB, c2->mode)) return 18; return 0; /* shouldn't get here */ } @@ -417,8 +354,6 @@ int codec2_samples_per_frame(struct CODEC2 *c2) { if (CODEC2_MODE_ACTIVE(CODEC2_MODE_1300, c2->mode)) return 320; if (CODEC2_MODE_ACTIVE(CODEC2_MODE_1200, c2->mode)) return 320; if (CODEC2_MODE_ACTIVE(CODEC2_MODE_700C, c2->mode)) return 320; - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450, c2->mode)) return 320; - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450PWB, c2->mode)) return 640; return 0; /* shouldn't get here */ } @@ -1640,26 +1575,6 @@ float codec2_energy_700c(struct CODEC2 *c2, const unsigned char *bits) { return POW10F(mean / 10.0); } -float codec2_energy_450(struct CODEC2 *c2, const unsigned char *bits) { - int indexes[4]; - unsigned int nbit = 0; - - assert(c2 != NULL); - - /* unpack bits from channel ------------------------------------*/ - - indexes[0] = unpack_natural_or_gray(bits, &nbit, 9, 0); - // indexes[1] = unpack_natural_or_gray(bits, &nbit, 9, 0); - indexes[2] = unpack_natural_or_gray(bits, &nbit, 3, 0); - indexes[3] = unpack_natural_or_gray(bits, &nbit, 6, 0); - - float mean = newamp2_energy_cb[0].cb[indexes[2]]; - mean -= 10; - if (indexes[3] == 0) mean -= 10; - - return POW10F(mean / 10.0); -} - /*---------------------------------------------------------------------------*\ FUNCTION....: codec2_get_energy() @@ -1678,9 +1593,7 @@ float codec2_get_energy(struct CODEC2 *c2, const unsigned char *bits) { (CODEC2_MODE_ACTIVE(CODEC2_MODE_1400, c2->mode)) || (CODEC2_MODE_ACTIVE(CODEC2_MODE_1300, c2->mode)) || (CODEC2_MODE_ACTIVE(CODEC2_MODE_1200, c2->mode)) || - (CODEC2_MODE_ACTIVE(CODEC2_MODE_700C, c2->mode)) || - (CODEC2_MODE_ACTIVE(CODEC2_MODE_450, c2->mode)) || - (CODEC2_MODE_ACTIVE(CODEC2_MODE_450PWB, c2->mode))); + (CODEC2_MODE_ACTIVE(CODEC2_MODE_700C, c2->mode))); MODEL model; float xq_dec[2] = {}; int e_index, WoE_index; @@ -1720,203 +1633,10 @@ float codec2_get_energy(struct CODEC2 *c2, const unsigned char *bits) { if (CODEC2_MODE_ACTIVE(CODEC2_MODE_700C, c2->mode)) { e = codec2_energy_700c(c2, bits); } - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_450, c2->mode) || - CODEC2_MODE_ACTIVE(CODEC2_MODE_450PWB, c2->mode)) { - e = codec2_energy_450(c2, bits); - } return e; } -/*---------------------------------------------------------------------------*\ - - FUNCTION....: codec2_encode_450 - AUTHOR......: Thomas Kurin and Stefan Erhardt - INSTITUTE...: Institute for Electronics Engineering, University of -Erlangen-Nuremberg DATE CREATED: July 2018 - - 450 bit/s codec that uses newamp2 fixed rate VQ of amplitudes. - - Encodes 320 speech samples (40ms of speech) into 28 bits. - - The codec2 algorithm actually operates internally on 10ms (80 - sample) frames, so we run the encoding algorithm four times: - - frame 0: nothing - frame 1: nothing - frame 2: nothing - frame 3: 9 bit 1 stage VQ, 3 bits energy, - 6 bit scalar Wo/voicing/plosive. No spare bits. - - If a plosive is detected the frame at the energy-step is encoded. - - Voicing is encoded using the 000000 index of the Wo quantiser. - Plosive is encoded using the 111111 index of the Wo quantiser. - - The bit allocation is: - - Parameter frames 1-3 frame 4 Total - ----------------------------------------------------------- - Harmonic magnitudes (rate k VQ) 0 9 9 - Energy 0 3 3 - log Wo/voicing/plosive 0 6 6 - TOTAL 0 18 18 - - -\*---------------------------------------------------------------------------*/ - -void codec2_encode_450(struct CODEC2 *c2, unsigned char *bits, short speech[]) { - MODEL model; - int indexes[4], i, h, M = 4; - unsigned int nbit = 0; - int plosiv = 0; - float energydelta[M]; - int spectralCounter; - - assert(c2 != NULL); - - memset(bits, '\0', ((codec2_bits_per_frame(c2) + 7) / 8)); - for (i = 0; i < M; i++) { - analyse_one_frame(c2, &model, &speech[i * c2->n_samp]); - energydelta[i] = 0; - spectralCounter = 0; - for (h = 0; h < (model.L); h++) { - // only detect above 300 Hz - if (h * model.Wo * (c2->c2const.Fs / 2000.0) / M_PI > 0.3) { - energydelta[i] = - (double)energydelta[i] + (double)20.0 * log10(model.A[10] + 1E-16); - spectralCounter = spectralCounter + 1; - } - } - energydelta[i] = energydelta[i] / spectralCounter; - } - // Constants for plosive Detection tdB = threshold; minPwr = from below this - // level plosives have to rise - float tdB = 15; // not fixed can be changed - float minPwr = 15; // not fixed can be changed - if ((c2->energy_prev) < minPwr && - energydelta[0] > ((c2->energy_prev) + tdB)) { - plosiv = 1; - } - if (energydelta[0] < minPwr && energydelta[1] > (energydelta[0] + tdB)) { - plosiv = 2; - } - if (energydelta[1] < minPwr && energydelta[2] > (energydelta[1] + tdB)) { - plosiv = 3; - } - if (energydelta[2] < minPwr && energydelta[3] > (energydelta[2] + tdB)) { - plosiv = 4; - } - if (plosiv != 0 && plosiv != 4) { - analyse_one_frame(c2, &model, &speech[(plosiv - 1) * c2->n_samp]); - } - - c2->energy_prev = energydelta[3]; - - int K = 29; - float rate_K_vec[K], mean; - float rate_K_vec_no_mean[K], rate_K_vec_no_mean_[K]; - if (plosiv > 0) { - plosiv = 1; - } - newamp2_model_to_indexes(&c2->c2const, indexes, &model, rate_K_vec, - c2->n2_rate_K_sample_freqs_kHz, K, &mean, - rate_K_vec_no_mean, rate_K_vec_no_mean_, plosiv); - - pack_natural_or_gray(bits, &nbit, indexes[0], 9, 0); - // pack_natural_or_gray(bits, &nbit, indexes[1], 9, 0); - pack_natural_or_gray(bits, &nbit, indexes[2], 3, 0); - pack_natural_or_gray(bits, &nbit, indexes[3], 6, 0); - - assert(nbit == (unsigned)codec2_bits_per_frame(c2)); -} - -/*---------------------------------------------------------------------------*\ - - FUNCTION....: codec2_decode_450 - AUTHOR......: Thomas Kurin and Stefan Erhardt - INSTITUTE...: Institute for Electronics Engineering, University of -Erlangen-Nuremberg DATE CREATED: July 2018 - -\*---------------------------------------------------------------------------*/ - -void codec2_decode_450(struct CODEC2 *c2, short speech[], - const unsigned char *bits) { - MODEL model[4]; - int indexes[4]; - int i; - unsigned int nbit = 0; - - assert(c2 != NULL); - - /* unpack bits from channel ------------------------------------*/ - - indexes[0] = unpack_natural_or_gray(bits, &nbit, 9, 0); - // indexes[1] = unpack_natural_or_gray(bits, &nbit, 9, 0); - indexes[2] = unpack_natural_or_gray(bits, &nbit, 3, 0); - indexes[3] = unpack_natural_or_gray(bits, &nbit, 6, 0); - - int M = 4; - COMP HH[M][MAX_AMP + 1]; - float interpolated_surface_[M][NEWAMP2_K]; - int pwbFlag = 0; - - newamp2_indexes_to_model( - &c2->c2const, model, (COMP *)HH, (float *)interpolated_surface_, - c2->n2_prev_rate_K_vec_, &c2->Wo_left, &c2->voicing_left, - c2->n2_rate_K_sample_freqs_kHz, NEWAMP2_K, c2->phase_fft_fwd_cfg, - c2->phase_fft_inv_cfg, indexes, 1.5, pwbFlag); - - for (i = 0; i < M; i++) { - synthesise_one_frame(c2, &speech[c2->n_samp * i], &model[i], &HH[i][0], - 1.5); - } -} - -/*---------------------------------------------------------------------------*\ - - FUNCTION....: codec2_decode_450pwb - AUTHOR......: Thomas Kurin and Stefan Erhardt - INSTITUTE...: Institute for Electronics Engineering, University of -Erlangen-Nuremberg DATE CREATED: July 2018 - - Decodes the 450 codec data in pseudo wideband at 16kHz samplerate. - -\*---------------------------------------------------------------------------*/ - -void codec2_decode_450pwb(struct CODEC2 *c2, short speech[], - const unsigned char *bits) { - MODEL model[4]; - int indexes[4]; - int i; - unsigned int nbit = 0; - - assert(c2 != NULL); - - /* unpack bits from channel ------------------------------------*/ - - indexes[0] = unpack_natural_or_gray(bits, &nbit, 9, 0); - // indexes[1] = unpack_natural_or_gray(bits, &nbit, 9, 0); - indexes[2] = unpack_natural_or_gray(bits, &nbit, 3, 0); - indexes[3] = unpack_natural_or_gray(bits, &nbit, 6, 0); - - int M = 4; - COMP HH[M][MAX_AMP + 1]; - float interpolated_surface_[M][NEWAMP2_16K_K]; - int pwbFlag = 1; - - newamp2_indexes_to_model( - &c2->c2const, model, (COMP *)HH, (float *)interpolated_surface_, - c2->n2_pwb_prev_rate_K_vec_, &c2->Wo_left, &c2->voicing_left, - c2->n2_pwb_rate_K_sample_freqs_kHz, NEWAMP2_16K_K, c2->phase_fft_fwd_cfg, - c2->phase_fft_inv_cfg, indexes, 1.5, pwbFlag); - - for (i = 0; i < M; i++) { - synthesise_one_frame(c2, &speech[c2->n_samp * i], &model[i], &HH[i][0], - 1.5); - } -} - /*---------------------------------------------------------------------------* \ FUNCTION....: synthesise_one_frame() @@ -1931,10 +1651,8 @@ void synthesise_one_frame(struct CODEC2 *c2, short speech[], MODEL *model, COMP Aw[], float gain) { int i; - if (CODEC2_MODE_ACTIVE(CODEC2_MODE_700C, c2->mode) || - CODEC2_MODE_ACTIVE(CODEC2_MODE_450, c2->mode) || - CODEC2_MODE_ACTIVE(CODEC2_MODE_450PWB, c2->mode)) { - /* newamp1/2, we've already worked out rate L phase */ + if (CODEC2_MODE_ACTIVE(CODEC2_MODE_700C, c2->mode)) { + /* newamp1, we've already worked out rate L phase */ COMP *H = Aw; phase_synth_zero_order(c2->n_samp, model, &c2->ex_phase, H); } else { diff --git a/src/codec2.h b/src/codec2.h index 892a0dac..9f88264c 100644 --- a/src/codec2.h +++ b/src/codec2.h @@ -42,8 +42,6 @@ extern "C" { #define CODEC2_MODE_1300 4 #define CODEC2_MODE_1200 5 #define CODEC2_MODE_700C 8 -#define CODEC2_MODE_450 10 -#define CODEC2_MODE_450PWB 11 #ifndef CODEC2_MODE_EN_DEFAULT #define CODEC2_MODE_EN_DEFAULT 1 @@ -78,12 +76,6 @@ extern "C" { #if !defined(CODEC2_MODE_700C_EN) #define CODEC2_MODE_700C_EN CODEC2_MODE_EN_DEFAULT #endif -#if !defined(CODEC2_MODE_450_EN) -#define CODEC2_MODE_450_EN CODEC2_MODE_EN_DEFAULT -#endif -#if !defined(CODEC2_MODE_450PWB_EN) -#define CODEC2_MODE_450PWB_EN CODEC2_MODE_EN_DEFAULT -#endif #define CODEC2_MODE_ACTIVE(mode_name, var) \ ((mode_name##_EN) == 0 ? 0 : (var) == mode_name) diff --git a/src/newamp2.c b/src/newamp2.c deleted file mode 100644 index 0b5f0322..00000000 --- a/src/newamp2.c +++ /dev/null @@ -1,536 +0,0 @@ -/*---------------------------------------------------------------------------*\ - - 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 "newamp2.h" - -#include -#include -#include -#include -#include - -#include "defines.h" -#include "mbest.h" -#include "newamp1.h" -#include "phase.h" -#include "quantise.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; k < K; k++) { - rate_K_sample_freqs_kHz[k] = freq[k]; - // printf("%f ",mel); - // printf("%f \n",rate_K_sample_freqs_kHz[k]); - } -} - -/*---------------------------------------------------------------------------*\ - - FUNCTION....: n2_resample_const_rate_f() still equal to -resample_const_rate_f() AUTHOR......: David Rowe DATE CREATED: Jan 2017 - - Resample Am from time-varying rate L=floor(pi/Wo) to fixed rate K. - -\*---------------------------------------------------------------------------*/ - -void n2_resample_const_rate_f(C2CONST *c2const, MODEL *model, - float rate_K_vec[], - float rate_K_sample_freqs_kHz[], int K) { - int m; - float AmdB[MAX_AMP + 1], rate_L_sample_freqs_kHz[MAX_AMP + 1], AmdB_peak; - - /* convert rate L=pi/Wo amplitude samples to fixed rate K */ - - AmdB_peak = -100.0; - for (m = 1; m <= model->L; 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; i < ndim; i++) w[i] = 1.0; - - mbest_stage1 = mbest_create(1); - - index[0] = 0; - - /* Stage 1 */ - - mbest_search450(codebook1, x, w, ndim, NEWAMP2_K, newamp2vq_cb[0].m, - mbest_stage1, index); - n1 = mbest_stage1->list[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; k < K; k++) { - rate_K_vec_term[k + 1] = rate_K_vec[k]; - rate_K_sample_freqs_kHz_term[k + 1] = rate_K_sample_freqs_kHz[k]; - - // printf("k: %d f: %f rate_K: %f\n", k, rate_K_sample_freqs_kHz[k], - // rate_K_vec[k]); - } - - for (m = 1; m <= model->L; 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; k < K; k++) { - pre[k] = 20.0 * log10f(sample_freq_kHz[k] / 0.3); - vec[k] += pre[k]; - e_before += POW10F(vec[k] / 10.0); - vec[k] *= pf_gain; - e_after += POW10F(vec[k] / 10.0); - } - - float gain = e_after / e_before; - float gaindB = 10 * log10f(gain); - - for (k = 0; k < K; k++) { - vec[k] -= gaindB; - vec[k] -= pre[k]; - } -} - -/*---------------------------------------------------------------------------*\ - - FUNCTION....: newamp2_model_to_indexes - AUTHOR......: Thomas Kurin and Stefan Erhardt - INSTITUTE...: Institute for Electronics Engineering, University of -Erlangen-Nuremberg DATE CREATED: July 2018 - - newamp2 encoder: Encodes the 8k sampled samples using mbest search (one stage) - -\*---------------------------------------------------------------------------*/ - -void newamp2_model_to_indexes(C2CONST *c2const, int indexes[], MODEL *model, - float rate_K_vec[], - float rate_K_sample_freqs_kHz[], int K, - float *mean, float rate_K_vec_no_mean[], - float rate_K_vec_no_mean_[], int plosive) { - int k; - - /* convert variable rate L to fixed rate K */ - - resample_const_rate_f(c2const, model, rate_K_vec, rate_K_sample_freqs_kHz, K); - - /* remove mean and two stage VQ */ - - float sum = 0.0; - for (k = 0; k < K; k++) sum += rate_K_vec[k]; - *mean = sum / K; - for (k = 0; k < K; k++) { - rate_K_vec_no_mean[k] = rate_K_vec[k] - *mean; - } - // NEWAMP2_16K_K+1 because the last vector is not a vector for VQ (and not - // included in the constant) but a calculated medium mean value - n2_rate_K_mbest_encode(indexes, rate_K_vec_no_mean, rate_K_vec_no_mean_, - NEWAMP2_16K_K + 1); - - /* scalar quantise mean (effectively the frame energy) */ - - float w[1] = {1.0}; - float se; - indexes[2] = quantise(newamp2_energy_cb[0].cb, mean, w, - newamp2_energy_cb[0].k, newamp2_energy_cb[0].m, &se); - - /* scalar quantise Wo. We steal the smallest Wo index to signal - an unvoiced frame */ - - if (model->voiced) { - 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; k < K; k++) { - rate_K_vec_no_mean_[k] = codebook1[(NEWAMP2_16K_K + 1) * n1 + k]; - } - - post_filter_newamp1(rate_K_vec_no_mean_, rate_K_sample_freqs_kHz, K, pf_gain); - - *mean_ = newamp2_energy_cb[0].cb[indexes[2]]; - - for (k = 0; k < K; k++) { - rate_K_vec_[k] = rate_K_vec_no_mean_[k] + *mean_; - } -} - -/*---------------------------------------------------------------------------*\ - - FUNCTION....: newamp2_16k_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. Extends the sample rate by looking up the -corresponding higher frequency values with their energy difference to the base -energy (=>mean2) - -\*---------------------------------------------------------------------------*/ - -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; k < K; k++) { - rate_K_vec_no_mean_[k] = codebook1[(K + 1) * n1 + k]; - } - - n2_post_filter_newamp2(rate_K_vec_no_mean_, rate_K_sample_freqs_kHz, K, - pf_gain); - - *mean_ = newamp2_energy_cb[0].cb[indexes[2]]; - mean2 = *mean_ + codebook1[(K + 1) * n1 + K] - 10; - - // HF ear Protection - if (mean2 > 50) { - mean2 = 50; - } - - for (k = 0; k < K; k++) { - if (k < NEWAMP2_K) { - rate_K_vec_[k] = rate_K_vec_no_mean_[k] + *mean_; - } else { - // Amplify or Reduce ?? - rate_K_vec_[k] = rate_K_vec_no_mean_[k] + mean2; - } - } -} -/*---------------------------------------------------------------------------*\ - - FUNCTION....: newamp2_interpolate - AUTHOR......: Thomas Kurin and Stefan Erhardt - INSTITUTE...: Institute for Electronics Engineering, University of -Erlangen-Nuremberg DATE CREATED: July 2018 - - Interpolates to the 4 10ms Frames and leaves the first 2 empty for plosives - -\*---------------------------------------------------------------------------*/ - -void newamp2_interpolate(float interpolated_surface_[], float left_vec[], - float right_vec[], int K, int plosive_flag) { - int i, k; - int M = 4; - float c; - - /* (linearly) interpolate 25Hz amplitude vectors back to 100Hz */ - - if (plosive_flag == 0) { - for (i = 0, c = 1.0; i < M; i++, c -= 1.0 / M) { - for (k = 0; k < K; k++) { - interpolated_surface_[i * K + k] = - left_vec[k] * c + right_vec[k] * (1.0 - c); - } - } - } else { - for (i = 0, c = 1.0; i < M; i++, c -= 1.0 / M) { - for (k = 0; k < K; k++) { - if (i < 2) { - interpolated_surface_[i * K + k] = 0; - } else { - // perhaps add some dB ? - interpolated_surface_[i * K + k] = right_vec[k]; - } - } - } - } -} - -/*---------------------------------------------------------------------------*\ - - FUNCTION....: newamp2_indexes_to_model - AUTHOR......: Thomas Kurin and Stefan Erhardt - INSTITUTE...: Institute for Electronics Engineering, University of -Erlangen-Nuremberg DATE CREATED: July 2018 - - newamp2 decoder. Chooses whether to decode to 16k mode or to 8k mode - -\*---------------------------------------------------------------------------*/ - -void newamp2_indexes_to_model(C2CONST *c2const, MODEL model_[], COMP H[], - float *interpolated_surface_, - float prev_rate_K_vec_[], float *Wo_left, - int *voicing_left, - float rate_K_sample_freqs_kHz[], int K, - codec2_fft_cfg fwd_cfg, codec2_fft_cfg inv_cfg, - int indexes[], float pf_gain, int flag16k) { - float rate_K_vec_[K], rate_K_vec_no_mean_[K], mean_, Wo_right; - int voicing_right, k; - int M = 4; - - /* extract latest rate K vector */ - - if (flag16k == 0) { - newamp2_indexes_to_rate_K_vec(rate_K_vec_, rate_K_vec_no_mean_, - rate_K_sample_freqs_kHz, K, &mean_, indexes, - pf_gain); - } else { - newamp2_16k_indexes_to_rate_K_vec(rate_K_vec_, rate_K_vec_no_mean_, - rate_K_sample_freqs_kHz, K, &mean_, - indexes, pf_gain); - } - - /* decode latest Wo and voicing and plosive */ - int plosive_flag = 0; - - // Voiced with Wo - if (indexes[3] > 0 && 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; i < M; i++) { - model_[i].Wo = aWo_[i]; - model_[i].L = aL_[i]; - model_[i].voiced = avoicing_[i]; - // Plosive Detected - if (plosive_flag > 0) { - // 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 < K; k++) { - prev_rate_K_vec_[k] = rate_K_vec_[k]; - } - *Wo_left = Wo_right; - *voicing_left = voicing_right; -}