dsputil: Split audio operations off into a separate context
[libav.git] / libavcodec / acelp_pitch_delay.h
1 /*
2 * gain code, gain pitch and pitch delay decoding
3 *
4 * Copyright (c) 2008 Vladimir Voroshilov
5 *
6 * This file is part of Libav.
7 *
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 #ifndef AVCODEC_ACELP_PITCH_DELAY_H
24 #define AVCODEC_ACELP_PITCH_DELAY_H
25
26 #include <stdint.h>
27
28 #include "audiodsp.h"
29
30 #define PITCH_DELAY_MIN 20
31 #define PITCH_DELAY_MAX 143
32
33 /**
34 * @brief Decode pitch delay of the first subframe encoded by 8 bits with 1/3
35 * resolution.
36 * @param ac_index adaptive codebook index (8 bits)
37 *
38 * @return pitch delay in 1/3 units
39 *
40 * Pitch delay is coded:
41 * with 1/3 resolution, 19 < pitch_delay < 85
42 * integers only, 85 <= pitch_delay <= 143
43 */
44 int ff_acelp_decode_8bit_to_1st_delay3(int ac_index);
45
46 /**
47 * @brief Decode pitch delay of the second subframe encoded by 5 or 6 bits
48 * with 1/3 precision.
49 * @param ac_index adaptive codebook index (5 or 6 bits)
50 * @param pitch_delay_min lower bound (integer) of pitch delay interval
51 * for second subframe
52 *
53 * @return pitch delay in 1/3 units
54 *
55 * Pitch delay is coded:
56 * with 1/3 resolution, -6 < pitch_delay - int(prev_pitch_delay) < 5
57 *
58 * @remark The routine is used in G.729 @@8k, AMR @@10.2k, AMR @@7.95k,
59 * AMR @@7.4k for the second subframe.
60 */
61 int ff_acelp_decode_5_6_bit_to_2nd_delay3(
62 int ac_index,
63 int pitch_delay_min);
64
65 /**
66 * @brief Decode pitch delay with 1/3 precision.
67 * @param ac_index adaptive codebook index (4 bits)
68 * @param pitch_delay_min lower bound (integer) of pitch delay interval for
69 * second subframe
70 *
71 * @return pitch delay in 1/3 units
72 *
73 * Pitch delay is coded:
74 * integers only, -6 < pitch_delay - int(prev_pitch_delay) <= -2
75 * with 1/3 resolution, -2 < pitch_delay - int(prev_pitch_delay) < 1
76 * integers only, 1 <= pitch_delay - int(prev_pitch_delay) < 5
77 *
78 * @remark The routine is used in G.729 @@6.4k, AMR @@6.7k, AMR @@5.9k,
79 * AMR @@5.15k, AMR @@4.75k for the second subframe.
80 */
81 int ff_acelp_decode_4bit_to_2nd_delay3(
82 int ac_index,
83 int pitch_delay_min);
84
85 /**
86 * @brief Decode pitch delay of the first subframe encoded by 9 bits
87 * with 1/6 precision.
88 * @param ac_index adaptive codebook index (9 bits)
89 *
90 * @return pitch delay in 1/6 units
91 *
92 * Pitch delay is coded:
93 * with 1/6 resolution, 17 < pitch_delay < 95
94 * integers only, 95 <= pitch_delay <= 143
95 *
96 * @remark The routine is used in AMR @@12.2k for the first and third subframes.
97 */
98 int ff_acelp_decode_9bit_to_1st_delay6(int ac_index);
99
100 /**
101 * @brief Decode pitch delay of the second subframe encoded by 6 bits
102 * with 1/6 precision.
103 * @param ac_index adaptive codebook index (6 bits)
104 * @param pitch_delay_min lower bound (integer) of pitch delay interval for
105 * second subframe
106 *
107 * @return pitch delay in 1/6 units
108 *
109 * Pitch delay is coded:
110 * with 1/6 resolution, -6 < pitch_delay - int(prev_pitch_delay) < 5
111 *
112 * @remark The routine is used in AMR @@12.2k for the second and fourth subframes.
113 */
114 int ff_acelp_decode_6bit_to_2nd_delay6(
115 int ac_index,
116 int pitch_delay_min);
117
118 /**
119 * @brief Update past quantized energies
120 * @param[in,out] quant_energy past quantized energies (5.10)
121 * @param gain_corr_factor gain correction factor
122 * @param log2_ma_pred_order log2() of MA prediction order
123 * @param erasure frame erasure flag
124 *
125 * If frame erasure flag is not equal to zero, memory is updated with
126 * averaged energy, attenuated by 4dB:
127 * max(avg(quant_energy[i])-4, -14), i=0,ma_pred_order
128 *
129 * In normal mode memory is updated with
130 * Er - Ep = 20 * log10(gain_corr_factor)
131 *
132 * @remark The routine is used in G.729 and AMR (all modes).
133 */
134 void ff_acelp_update_past_gain(
135 int16_t* quant_energy,
136 int gain_corr_factor,
137 int log2_ma_pred_order,
138 int erasure);
139
140 /**
141 * @brief Decode the adaptive codebook gain and add
142 * correction (4.1.5 and 3.9.1 of G.729).
143 * @param adsp initialized audio DSP context
144 * @param gain_corr_factor gain correction factor (2.13)
145 * @param fc_v fixed-codebook vector (2.13)
146 * @param mr_energy mean innovation energy and fixed-point correction (7.13)
147 * @param[in,out] quant_energy past quantized energies (5.10)
148 * @param subframe_size length of subframe
149 *
150 * @return quantized fixed-codebook gain (14.1)
151 *
152 * The routine implements equations 69, 66 and 71 of the G.729 specification (3.9.1)
153 *
154 * Em - mean innovation energy (dB, constant, depends on decoding algorithm)
155 * Ep - mean-removed predicted energy (dB)
156 * Er - mean-removed innovation energy (dB)
157 * Ei - mean energy of the fixed-codebook contribution (dB)
158 * N - subframe_size
159 * M - MA (Moving Average) prediction order
160 * gc - fixed-codebook gain
161 * gc_p - predicted fixed-codebook gain
162 *
163 * Fixed codebook gain is computed using predicted gain gc_p and
164 * correction factor gain_corr_factor as shown below:
165 *
166 * gc = gc_p * gain_corr_factor
167 *
168 * The predicted fixed codebook gain gc_p is found by predicting
169 * the energy of the fixed-codebook contribution from the energy
170 * of previous fixed-codebook contributions.
171 *
172 * mean = 1/N * sum(i,0,N){ fc_v[i] * fc_v[i] }
173 *
174 * Ei = 10log(mean)
175 *
176 * Er = 10log(1/N * gc^2 * mean) - Em = 20log(gc) + Ei - Em
177 *
178 * Replacing Er with Ep and gc with gc_p we will receive:
179 *
180 * Ep = 10log(1/N * gc_p^2 * mean) - Em = 20log(gc_p) + Ei - Em
181 *
182 * and from above:
183 *
184 * gc_p = 10^((Ep - Ei + Em) / 20)
185 *
186 * Ep is predicted using past energies and prediction coefficients:
187 *
188 * Ep = sum(i,0,M){ ma_prediction_coeff[i] * quant_energy[i] }
189 *
190 * gc_p in fixed-point arithmetic is calculated as following:
191 *
192 * mean = 1/N * sum(i,0,N){ (fc_v[i] / 2^13) * (fc_v[i] / 2^13) } =
193 * = 1/N * sum(i,0,N) { fc_v[i] * fc_v[i] } / 2^26
194 *
195 * Ei = 10log(mean) = -10log(N) - 10log(2^26) +
196 * + 10log(sum(i,0,N) { fc_v[i] * fc_v[i] })
197 *
198 * Ep - Ei + Em = Ep + Em + 10log(N) + 10log(2^26) -
199 * - 10log(sum(i,0,N) { fc_v[i] * fc_v[i] }) =
200 * = Ep + mr_energy - 10log(sum(i,0,N) { fc_v[i] * fc_v[i] })
201 *
202 * gc_p = 10 ^ ((Ep - Ei + Em) / 20) =
203 * = 2 ^ (3.3219 * (Ep - Ei + Em) / 20) = 2 ^ (0.166 * (Ep - Ei + Em))
204 *
205 * where
206 *
207 * mr_energy = Em + 10log(N) + 10log(2^26)
208 *
209 * @remark The routine is used in G.729 and AMR (all modes).
210 */
211 int16_t ff_acelp_decode_gain_code(
212 AudioDSPContext *adsp,
213 int gain_corr_factor,
214 const int16_t* fc_v,
215 int mr_energy,
216 const int16_t* quant_energy,
217 const int16_t* ma_prediction_coeff,
218 int subframe_size,
219 int max_pred_order);
220
221 /**
222 * Calculate fixed gain (part of section 6.1.3 of AMR spec)
223 *
224 * @param fixed_gain_factor gain correction factor
225 * @param fixed_mean_energy mean decoded algebraic codebook vector energy
226 * @param prediction_error vector of the quantified predictor errors of
227 * the four previous subframes. It is updated by this function.
228 * @param energy_mean desired mean innovation energy
229 * @param pred_table table of four moving average coefficients
230 */
231 float ff_amr_set_fixed_gain(float fixed_gain_factor, float fixed_mean_energy,
232 float *prediction_error, float energy_mean,
233 const float *pred_table);
234
235
236 /**
237 * Decode the adaptive codebook index to the integer and fractional parts
238 * of the pitch lag for one subframe at 1/3 fractional precision.
239 *
240 * The choice of pitch lag is described in 3GPP TS 26.090 section 5.6.1.
241 *
242 * @param lag_int integer part of pitch lag of the current subframe
243 * @param lag_frac fractional part of pitch lag of the current subframe
244 * @param pitch_index parsed adaptive codebook (pitch) index
245 * @param prev_lag_int integer part of pitch lag for the previous subframe
246 * @param subframe current subframe number
247 * @param third_as_first treat the third frame the same way as the first
248 */
249 void ff_decode_pitch_lag(int *lag_int, int *lag_frac, int pitch_index,
250 const int prev_lag_int, const int subframe,
251 int third_as_first, int resolution);
252
253 #endif /* AVCODEC_ACELP_PITCH_DELAY_H */