From e500315b1dd7654ec550b5fc013232b3567358e0 Mon Sep 17 00:00:00 2001 From: Vladimir Voroshilov Date: Mon, 30 Jun 2008 18:03:38 +0000 Subject: [PATCH] gain code, gain pitch and pitch delay decoding for ACELP based codecs Originally committed as revision 14037 to svn://svn.ffmpeg.org/ffmpeg/trunk --- libavcodec/acelp_pitch_delay.c | 119 ++++++++++++++++++ libavcodec/acelp_pitch_delay.h | 220 +++++++++++++++++++++++++++++++++ 2 files changed, 339 insertions(+) create mode 100644 libavcodec/acelp_pitch_delay.c create mode 100644 libavcodec/acelp_pitch_delay.h diff --git a/libavcodec/acelp_pitch_delay.c b/libavcodec/acelp_pitch_delay.c new file mode 100644 index 0000000000..9b35fcc892 --- /dev/null +++ b/libavcodec/acelp_pitch_delay.c @@ -0,0 +1,119 @@ +/* + * gain code, gain pitch and pitch delay decoding + * + * Copyright (c) 2008 Vladimir Voroshilov + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg 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 + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "avcodec.h" +#include "acelp_pitch_delay.h" +#include "acelp_math.h" + +int ff_acelp_decode_8bit_to_1st_delay3(int ac_index) +{ + ac_index += 58; + if(ac_index > 254) + ac_index = 3 * ac_index - 510; + return ac_index; +} + +int ff_acelp_decode_4bit_to_2nd_delay3( + int ac_index, + int pitch_delay_min) +{ + if(ac_index < 4) + return 3 * (ac_index + pitch_delay_min); + else if(ac_index < 12) + return 3 * pitch_delay_min + ac_index + 6; + else + return 3 * (ac_index + pitch_delay_min) - 18; +} + +int ff_acelp_decode_5_6_bit_to_2nd_delay3( + int ac_index, + int pitch_delay_min) +{ + return 3 * pitch_delay_min + ac_index - 2; +} + +int ff_acelp_decode_9bit_to_1st_delay6(int ac_index) +{ + if(ac_index < 463) + return ac_index + 105; + else + return 6 * (ac_index - 368); +} +int ff_acelp_decode_6bit_to_2nd_delay6( + int ac_index, + int pitch_delay_min) +{ + return 6 * pitch_delay_min + ac_index - 3; +} + +void ff_acelp_update_past_gain( + int16_t* quant_energy, + int gain_corr_factor, + int log2_ma_pred_order, + int erasure) +{ + int i; + int avg_gain=quant_energy[(1 << log2_ma_pred_order) - 1]; // (5.10) + + for(i=(1 << log2_ma_pred_order) - 1; i>0; i--) + { + avg_gain += quant_energy[i-1]; + quant_energy[i] = quant_energy[i-1]; + } + + if(erasure) + quant_energy[0] = FFMAX(avg_gain >> log2_ma_pred_order, -10240) - 4096; // -10 and -4 in (5.10) + else + quant_energy[0] = (6165 * ((ff_log2(gain_corr_factor) >> 2) - (13 << 13))) >> 13; +} + +int16_t ff_acelp_decode_gain_code( + int gain_corr_factor, + const int16_t* fc_v, + int mr_energy, + const int16_t* quant_energy, + const int16_t* ma_prediction_coeff, + int subframe_size, + int ma_pred_order) +{ + int i; + + mr_energy <<= 10; + + for(i=0; i> 3) & ~0x3ff); + + mr_energy = (5439 * (mr_energy >> 15)) >> 8; // (0.15) = (0.15) * (7.23) + + return bidir_sal( + ((ff_exp2(mr_energy & 0x7fff) + 16) >> 5) * (gain_corr_factor >> 1), + (mr_energy >> 15) - 25 + ); +#else + mr_energy = gain_corr_factor * exp(M_LN10 / (20 << 23) * mr_energy) / + sqrt(sum_of_squares(fc_v, subframe_size, 0, 0)); + return mr_energy >> 12; +#endif +} diff --git a/libavcodec/acelp_pitch_delay.h b/libavcodec/acelp_pitch_delay.h new file mode 100644 index 0000000000..b5b9add7f2 --- /dev/null +++ b/libavcodec/acelp_pitch_delay.h @@ -0,0 +1,220 @@ +/* + * gain code, gain pitch and pitch delay decoding + * + * Copyright (c) 2008 Vladimir Voroshilov + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg 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 + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef FFMPEG_ACELP_PITCH_DELAY_H +#define FFMPEG_ACELP_PITCH_DELAY_H + +#include + +#define PITCH_DELAY_MIN 20 +#define PITCH_DELAY_MAX 143 + +/** + * \brief Decode pitch delay of the first subframe encoded by 8 bits with 1/3 + * resolution. + * \param ac_index adaptive codebook index (8 bits) + * + * \return pitch delay in 1/3 units + * + * Pitch delay is coded: + * with 1/3 resolution, 19 < pitch_delay < 85 + * integers only, 85 <= pitch_delay <= 143 + */ +int ff_acelp_decode_8bit_to_1st_delay3(int ac_index); + +/** + * \brief Decode pitch delay of the second subframe encoded by 5 or 6 bits + * with 1/3 precision. + * \param ac_index adaptive codebook index (5 or 6 bits) + * \param pitch_delay_min lower bound (integer) of pitch delay interval + * for second subframe + * + * \return pitch delay in 1/3 units + * + * Pitch delay is coded: + * with 1/3 resolution, -6 < pitch_delay - int(prev_pitch_delay) < 5 + * + * \remark The routine is used in G.729 @8k, AMR @10.2k, AMR @7.95k, + * AMR @7.4k for the second subframe. + */ +int ff_acelp_decode_5_6_bit_to_2nd_delay3( + int ac_index, + int pitch_delay_min); + +/** + * \brief Decode pitch delay with 1/3 precision. + * \param ac_index adaptive codebook index (4 bits) + * \param pitch_delay_min lower bound (integer) of pitch delay interval for + * second subframe + * + * \return pitch delay in 1/3 units + * + * Pitch delay is coded: + * integers only, -6 < pitch_delay - int(prev_pitch_delay) <= -2 + * with 1/3 resolution, -2 < pitch_delay - int(prev_pitch_delay) < 1 + * integers only, 1 <= pitch_delay - int(prev_pitch_delay) < 5 + * + * \remark The routine is used in G.729 @6.4k, AMR @6.7k, AMR @5.9k, + * AMR @5.15k, AMR @4.75k for the second subframe. + */ +int ff_acelp_decode_4bit_to_2nd_delay3( + int ac_index, + int pitch_delay_min); + +/** + * \brief Decode pitch delay of the first subframe encoded by 9 bits + * with 1/6 precision. + * \param ac_index adaptive codebook index (9 bits) + * \param pitch_delay_min lower bound (integer) of pitch delay interval for + * second subframe + * + * \return pitch delay in 1/6 units + * + * Pitch delay is coded: + * with 1/6 resolution, 17 < pitch_delay < 95 + * integers only, 95 <= pitch_delay <= 143 + * + * \remark The routine is used in AMR @12.2k for the first and third subframes. + */ +int ff_acelp_decode_9bit_to_1st_delay6(int ac_index); + +/** + * \brief Decode pitch delay of the second subframe encoded by 6 bits + * with 1/6 precision. + * \param ac_index adaptive codebook index (6 bits) + * \param pitch_delay_min lower bound (integer) of pitch delay interval for + * second subframe + * + * \return pitch delay in 1/6 units + * + * Pitch delay is coded: + * with 1/6 resolution, -6 < pitch_delay - int(prev_pitch_delay) < 5 + * + * \remark The routine is used in AMR @12.2k for the second and fourth subframes. + */ +int ff_acelp_decode_6bit_to_2nd_delay6( + int ac_index, + int pitch_delay_min); + +/** + * \brief Update past quantized energies + * \param quant_energy [in/out] past quantized energies (5.10) + * \param gain_corr_factor gain correction factor + * \param log2_ma_pred_order log2() of MA prediction order + * \param erasure frame erasure flag + * + * If frame erasure flag is not equal to zero, memory is updated with + * averaged energy, attenuated by 4dB: + * max(avg(quant_energy[i])-4, -14), i=0,ma_pred_order + * + * In normal mode memory is updated with + * Er - Ep = 20 * log10(gain_corr_factor) + * + * \remark The routine is used in G.729 and AMR (all modes). + */ +void ff_acelp_update_past_gain( + int16_t* quant_energy, + int gain_corr_factor, + int log2_ma_pred_order, + int erasure); + +/** + * \brief Decode the adaptive codebook gain and add + * correction (4.1.5 and 3.9.1 of G.729). + * \param gain_corr_factor gain correction factor (2.13) + * \param fc_v fixed-codebook vector (2.13) + * \param mr_energy mean innovation energy and fixed-point correction (7.13) + * \param quant_energy [in/out] past quantized energies (5.10) + * \param subframe_size length of subframe + * \param ma_pred_order MA prediction order + * + * \return quantized fixed-codebook gain (14.1) + * + * The routine implements equations 69, 66 and 71 of the G.729 specification (3.9.1) + * + * Em - mean innovation energy (dB, constant, depends on decoding algorithm) + * Ep - mean-removed predicted energy (dB) + * Er - mean-removed innovation energy (dB) + * Ei - mean energy of the fixed-codebook contribution (dB) + * N - subframe_size + * M - MA (Moving Average) prediction order + * gc - fixed-codebook gain + * gc_p - predicted fixed-codebook gain + * + * Fixed codebook gain is computed using predicted gain gc_p and + * correction factor gain_corr_factor as shown below: + * + * gc = gc_p * gain_corr_factor + * + * The predicted fixed codebook gain gc_p is found by predicting + * the energy of the fixed-codebook contribution from the energy + * of previous fixed-codebook contributions. + * + * mean = 1/N * sum(i,0,N){ fc_v[i] * fc_v[i] } + * + * Ei = 10log(mean) + * + * Er = 10log(1/N * gc^2 * mean) - Em = 20log(gc) + Ei - Em + * + * Replacing Er with Ep and gc with gc_p we will receive: + * + * Ep = 10log(1/N * gc_p^2 * mean) - Em = 20log(gc_p) + Ei - Em + * + * and from above: + * + * gc_p = 10^((Ep - Ei + Em) / 20) + * + * Ep is predicted using past energies and prediction coefficients: + * + * Ep = sum(i,0,M){ ma_prediction_coeff[i] * quant_energy[i] } + * + * gc_p in fixed-point arithmetic is calculated as following: + * + * mean = 1/N * sum(i,0,N){ (fc_v[i] / 2^13) * (fc_v[i] / 2^13) } = + * = 1/N * sum(i,0,N) { fc_v[i] * fc_v[i] } / 2^26 + * + * Ei = 10log(mean) = -10log(N) - 10log(2^26) + + * + 10log(sum(i,0,N) { fc_v[i] * fc_v[i] }) + * + * Ep - Ei + Em = Ep + Em + 10log(N) + 10log(2^26) - + * - 10log(sum(i,0,N) { fc_v[i] * fc_v[i] }) = + * = Ep + mr_energy - 10log(sum(i,0,N) { fc_v[i] * fc_v[i] }) + * + * gc_p = 10 ^ ((Ep - Ei + Em) / 20) = + * = 2 ^ (3.3219 * (Ep - Ei + Em) / 20) = 2 ^ (0.166 * (Ep - Ei + Em)) + * + * where + * + * mr_energy = Em + 10log(N) + 10log(2^26) + * + * \remark The routine is used in G.729 and AMR (all modes). + */ +int16_t ff_acelp_decode_gain_code( + int gain_corr_factor, + const int16_t* fc_v, + int mr_energy, + const int16_t* quant_energy, + const int16_t* ma_prediction_coeff, + int subframe_size, + int max_pred_order); + +#endif /* FFMPEG_ACELP_PITCH_DELAY_H */