1 /******************************************************************************
2 *
3 * Copyright 2022 Google LLC
4 *
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
8 *
9 * http://www.apache.org/licenses/LICENSE-2.0
10 *
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 *
17 ******************************************************************************/
18
19 #include "ltpf.h"
20 #include "tables.h"
21
22 #include "ltpf_neon.h"
23 #include "ltpf_arm.h"
24
25
26 /* ----------------------------------------------------------------------------
27 * Resampling
28 * -------------------------------------------------------------------------- */
29
30 /**
31 * Resampling coefficients
32 * The coefficients, in fixed Q15, are reordered by phase for each source
33 * samplerate (coefficient matrix transposed)
34 */
35
36 #ifndef resample_8k_12k8
37 static const int16_t h_8k_12k8_q15[8*10] = {
38 214, 417, -1052, -4529, 26233, -4529, -1052, 417, 214, 0,
39 180, 0, -1522, -2427, 24506, -5289, 0, 763, 156, -28,
40 92, -323, -1361, 0, 19741, -3885, 1317, 861, 0, -61,
41 0, -457, -752, 1873, 13068, 0, 2389, 598, -213, -79,
42 -61, -398, 0, 2686, 5997, 5997, 2686, 0, -398, -61,
43 -79, -213, 598, 2389, 0, 13068, 1873, -752, -457, 0,
44 -61, 0, 861, 1317, -3885, 19741, 0, -1361, -323, 92,
45 -28, 156, 763, 0, -5289, 24506, -2427, -1522, 0, 180,
46 };
47 #endif /* resample_8k_12k8 */
48
49 #ifndef resample_16k_12k8
50 static const int16_t h_16k_12k8_q15[4*20] = {
51 -61, 214, -398, 417, 0, -1052, 2686, -4529, 5997, 26233,
52 5997, -4529, 2686, -1052, 0, 417, -398, 214, -61, 0,
53
54 -79, 180, -213, 0, 598, -1522, 2389, -2427, 0, 24506,
55 13068, -5289, 1873, 0, -752, 763, -457, 156, 0, -28,
56
57 -61, 92, 0, -323, 861, -1361, 1317, 0, -3885, 19741,
58 19741, -3885, 0, 1317, -1361, 861, -323, 0, 92, -61,
59
60 -28, 0, 156, -457, 763, -752, 0, 1873, -5289, 13068,
61 24506, 0, -2427, 2389, -1522, 598, 0, -213, 180, -79,
62 };
63 #endif /* resample_16k_12k8 */
64
65 #ifndef resample_32k_12k8
66 static const int16_t h_32k_12k8_q15[2*40] = {
67 -30, -31, 46, 107, 0, -199, -162, 209, 430, 0,
68 -681, -526, 658, 1343, 0, -2264, -1943, 2999, 9871, 13116,
69 9871, 2999, -1943, -2264, 0, 1343, 658, -526, -681, 0,
70 430, 209, -162, -199, 0, 107, 46, -31, -30, 0,
71
72 -14, -39, 0, 90, 78, -106, -229, 0, 382, 299,
73 -376, -761, 0, 1194, 937, -1214, -2644, 0, 6534, 12253,
74 12253, 6534, 0, -2644, -1214, 937, 1194, 0, -761, -376,
75 299, 382, 0, -229, -106, 78, 90, 0, -39, -14,
76 };
77 #endif /* resample_32k_12k8 */
78
79 #ifndef resample_24k_12k8
80 static const int16_t h_24k_12k8_q15[8*30] = {
81 -50, 19, 143, -93, -290, 278, 485, -658, -701, 1396,
82 901, -3019, -1042, 10276, 17488, 10276, -1042, -3019, 901, 1396,
83 -701, -658, 485, 278, -290, -93, 143, 19, -50, 0,
84
85 -46, 0, 141, -45, -305, 185, 543, -501, -854, 1153,
86 1249, -2619, -1908, 8712, 17358, 11772, 0, -3319, 480, 1593,
87 -504, -796, 399, 367, -261, -142, 138, 40, -52, -5,
88
89 -41, -17, 133, 0, -304, 91, 574, -334, -959, 878,
90 1516, -2143, -2590, 7118, 16971, 13161, 1202, -3495, 0, 1731,
91 -267, -908, 287, 445, -215, -188, 125, 62, -52, -12,
92
93 -34, -30, 120, 41, -291, 0, 577, -164, -1015, 585,
94 1697, -1618, -3084, 5534, 16337, 14406, 2544, -3526, -523, 1800,
95 0, -985, 152, 509, -156, -230, 104, 83, -48, -19,
96
97 -26, -41, 103, 76, -265, -83, 554, 0, -1023, 288,
98 1791, -1070, -3393, 3998, 15474, 15474, 3998, -3393, -1070, 1791,
99 288, -1023, 0, 554, -83, -265, 76, 103, -41, -26,
100
101 -19, -48, 83, 104, -230, -156, 509, 152, -985, 0,
102 1800, -523, -3526, 2544, 14406, 16337, 5534, -3084, -1618, 1697,
103 585, -1015, -164, 577, 0, -291, 41, 120, -30, -34,
104
105 -12, -52, 62, 125, -188, -215, 445, 287, -908, -267,
106 1731, 0, -3495, 1202, 13161, 16971, 7118, -2590, -2143, 1516,
107 878, -959, -334, 574, 91, -304, 0, 133, -17, -41,
108
109 -5, -52, 40, 138, -142, -261, 367, 399, -796, -504,
110 1593, 480, -3319, 0, 11772, 17358, 8712, -1908, -2619, 1249,
111 1153, -854, -501, 543, 185, -305, -45, 141, 0, -46,
112 };
113 #endif /* resample_24k_12k8 */
114
115 #ifndef resample_48k_12k8
116 static const int16_t h_48k_12k8_q15[4*60] = {
117 -13, -25, -20, 10, 51, 71, 38, -47, -133, -145,
118 -42, 139, 277, 242, 0, -329, -511, -351, 144, 698,
119 895, 450, -535, -1510, -1697, -521, 1999, 5138, 7737, 8744,
120 7737, 5138, 1999, -521, -1697, -1510, -535, 450, 895, 698,
121 144, -351, -511, -329, 0, 242, 277, 139, -42, -145,
122 -133, -47, 38, 71, 51, 10, -20, -25, -13, 0,
123
124 -9, -23, -24, 0, 41, 71, 52, -23, -115, -152,
125 -78, 92, 254, 272, 76, -251, -493, -427, 0, 576,
126 900, 624, -262, -1309, -1763, -954, 1272, 4356, 7203, 8679,
127 8169, 5886, 2767, 0, -1542, -1660, -809, 240, 848, 796,
128 292, -252, -507, -398, -82, 199, 288, 183, 0, -130,
129 -145, -71, 20, 69, 60, 20, -15, -26, -17, -3,
130
131 -6, -20, -26, -8, 31, 67, 62, 0, -94, -152,
132 -108, 45, 223, 287, 143, -167, -454, -480, -134, 439,
133 866, 758, 0, -1071, -1748, -1295, 601, 3559, 6580, 8485,
134 8485, 6580, 3559, 601, -1295, -1748, -1071, 0, 758, 866,
135 439, -134, -480, -454, -167, 143, 287, 223, 45, -108,
136 -152, -94, 0, 62, 67, 31, -8, -26, -20, -6,
137
138 -3, -17, -26, -15, 20, 60, 69, 20, -71, -145,
139 -130, 0, 183, 288, 199, -82, -398, -507, -252, 292,
140 796, 848, 240, -809, -1660, -1542, 0, 2767, 5886, 8169,
141 8679, 7203, 4356, 1272, -954, -1763, -1309, -262, 624, 900,
142 576, 0, -427, -493, -251, 76, 272, 254, 92, -78,
143 -152, -115, -23, 52, 71, 41, 0, -24, -23, -9,
144 };
145 #endif /* resample_48k_12k8 */
146
147
148 /**
149 * High-pass 50Hz filtering, at 12.8 KHz samplerate
150 * hp50 Biquad filter state
151 * xn Input sample, in fixed Q30
152 * return Filtered sample, in fixed Q30
153 */
filter_hp50(struct lc3_ltpf_hp50_state * hp50,int32_t xn)154 LC3_HOT static inline int32_t filter_hp50(
155 struct lc3_ltpf_hp50_state *hp50, int32_t xn)
156 {
157 int32_t yn;
158
159 const int32_t a1 = -2110217691, a2 = 1037111617;
160 const int32_t b1 = -2110535566, b2 = 1055267782;
161
162 yn = (hp50->s1 + (int64_t)xn * b2) >> 30;
163 hp50->s1 = (hp50->s2 + (int64_t)xn * b1 - (int64_t)yn * a1);
164 hp50->s2 = ( (int64_t)xn * b2 - (int64_t)yn * a2);
165
166 return yn;
167 }
168
169 /**
170 * Resample from 8 / 16 / 32 KHz to 12.8 KHz Template
171 * p Resampling factor with compared to 192 KHz (8, 4 or 2)
172 * h Arrange by phase coefficients table
173 * hp50 High-Pass biquad filter state
174 * x [-d..-1] Previous, [0..ns-1] Current samples, Q15
175 * y, n [0..n-1] Output `n` processed samples, Q14
176 *
177 * The `x` vector is aligned on 32 bits
178 * The number of previous samples `d` accessed on `x` is :
179 * d: { 10, 20, 40 } - 1 for resampling factors 8, 4 and 2.
180 */
181 #if !defined(resample_8k_12k8) || !defined(resample_16k_12k8) \
182 || !defined(resample_32k_12k8)
resample_x64k_12k8(const int p,const int16_t * h,struct lc3_ltpf_hp50_state * hp50,const int16_t * x,int16_t * y,int n)183 LC3_HOT static inline void resample_x64k_12k8(const int p, const int16_t *h,
184 struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
185 {
186 const int w = 2*(40 / p);
187
188 x -= w - 1;
189
190 for (int i = 0; i < 5*n; i += 5) {
191 const int16_t *hn = h + (i % p) * w;
192 const int16_t *xn = x + (i / p);
193 int32_t un = 0;
194
195 for (int k = 0; k < w; k += 10) {
196 un += *(xn++) * *(hn++);
197 un += *(xn++) * *(hn++);
198 un += *(xn++) * *(hn++);
199 un += *(xn++) * *(hn++);
200 un += *(xn++) * *(hn++);
201 un += *(xn++) * *(hn++);
202 un += *(xn++) * *(hn++);
203 un += *(xn++) * *(hn++);
204 un += *(xn++) * *(hn++);
205 un += *(xn++) * *(hn++);
206 }
207
208 int32_t yn = filter_hp50(hp50, un);
209 *(y++) = (yn + (1 << 15)) >> 16;
210 }
211 }
212 #endif
213
214 /**
215 * Resample from 24 / 48 KHz to 12.8 KHz Template
216 * p Resampling factor with compared to 192 KHz (8 or 4)
217 * h Arrange by phase coefficients table
218 * hp50 High-Pass biquad filter state
219 * x [-d..-1] Previous, [0..ns-1] Current samples, Q15
220 * y, n [0..n-1] Output `n` processed samples, Q14
221 *
222 * The `x` vector is aligned on 32 bits
223 * The number of previous samples `d` accessed on `x` is :
224 * d: { 30, 60 } - 1 for resampling factors 8 and 4.
225 */
226 #if !defined(resample_24k_12k8) || !defined(resample_48k_12k8)
resample_x192k_12k8(const int p,const int16_t * h,struct lc3_ltpf_hp50_state * hp50,const int16_t * x,int16_t * y,int n)227 LC3_HOT static inline void resample_x192k_12k8(const int p, const int16_t *h,
228 struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
229 {
230 const int w = 2*(120 / p);
231
232 x -= w - 1;
233
234 for (int i = 0; i < 15*n; i += 15) {
235 const int16_t *hn = h + (i % p) * w;
236 const int16_t *xn = x + (i / p);
237 int32_t un = 0;
238
239 for (int k = 0; k < w; k += 15) {
240 un += *(xn++) * *(hn++);
241 un += *(xn++) * *(hn++);
242 un += *(xn++) * *(hn++);
243 un += *(xn++) * *(hn++);
244 un += *(xn++) * *(hn++);
245 un += *(xn++) * *(hn++);
246 un += *(xn++) * *(hn++);
247 un += *(xn++) * *(hn++);
248 un += *(xn++) * *(hn++);
249 un += *(xn++) * *(hn++);
250 un += *(xn++) * *(hn++);
251 un += *(xn++) * *(hn++);
252 un += *(xn++) * *(hn++);
253 un += *(xn++) * *(hn++);
254 un += *(xn++) * *(hn++);
255 }
256
257 int32_t yn = filter_hp50(hp50, un);
258 *(y++) = (yn + (1 << 15)) >> 16;
259 }
260 }
261 #endif
262
263 /**
264 * Resample from 8 Khz to 12.8 KHz
265 * hp50 High-Pass biquad filter state
266 * x [-10..-1] Previous, [0..ns-1] Current samples, Q15
267 * y, n [0..n-1] Output `n` processed samples, Q14
268 *
269 * The `x` vector is aligned on 32 bits
270 */
271 #ifndef resample_8k_12k8
resample_8k_12k8(struct lc3_ltpf_hp50_state * hp50,const int16_t * x,int16_t * y,int n)272 LC3_HOT static void resample_8k_12k8(
273 struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
274 {
275 resample_x64k_12k8(8, h_8k_12k8_q15, hp50, x, y, n);
276 }
277 #endif /* resample_8k_12k8 */
278
279 /**
280 * Resample from 16 Khz to 12.8 KHz
281 * hp50 High-Pass biquad filter state
282 * x [-20..-1] Previous, [0..ns-1] Current samples, in fixed Q15
283 * y, n [0..n-1] Output `n` processed samples, in fixed Q14
284 *
285 * The `x` vector is aligned on 32 bits
286 */
287 #ifndef resample_16k_12k8
resample_16k_12k8(struct lc3_ltpf_hp50_state * hp50,const int16_t * x,int16_t * y,int n)288 LC3_HOT static void resample_16k_12k8(
289 struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
290 {
291 resample_x64k_12k8(4, h_16k_12k8_q15, hp50, x, y, n);
292 }
293 #endif /* resample_16k_12k8 */
294
295 /**
296 * Resample from 32 Khz to 12.8 KHz
297 * hp50 High-Pass biquad filter state
298 * x [-30..-1] Previous, [0..ns-1] Current samples, in fixed Q15
299 * y, n [0..n-1] Output `n` processed samples, in fixed Q14
300 *
301 * The `x` vector is aligned on 32 bits
302 */
303 #ifndef resample_32k_12k8
resample_32k_12k8(struct lc3_ltpf_hp50_state * hp50,const int16_t * x,int16_t * y,int n)304 LC3_HOT static void resample_32k_12k8(
305 struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
306 {
307 resample_x64k_12k8(2, h_32k_12k8_q15, hp50, x, y, n);
308 }
309 #endif /* resample_32k_12k8 */
310
311 /**
312 * Resample from 24 Khz to 12.8 KHz
313 * hp50 High-Pass biquad filter state
314 * x [-30..-1] Previous, [0..ns-1] Current samples, in fixed Q15
315 * y, n [0..n-1] Output `n` processed samples, in fixed Q14
316 *
317 * The `x` vector is aligned on 32 bits
318 */
319 #ifndef resample_24k_12k8
resample_24k_12k8(struct lc3_ltpf_hp50_state * hp50,const int16_t * x,int16_t * y,int n)320 LC3_HOT static void resample_24k_12k8(
321 struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
322 {
323 resample_x192k_12k8(8, h_24k_12k8_q15, hp50, x, y, n);
324 }
325 #endif /* resample_24k_12k8 */
326
327 /**
328 * Resample from 48 Khz to 12.8 KHz
329 * hp50 High-Pass biquad filter state
330 * x [-60..-1] Previous, [0..ns-1] Current samples, in fixed Q15
331 * y, n [0..n-1] Output `n` processed samples, in fixed Q14
332 *
333 * The `x` vector is aligned on 32 bits
334 */
335 #ifndef resample_48k_12k8
resample_48k_12k8(struct lc3_ltpf_hp50_state * hp50,const int16_t * x,int16_t * y,int n)336 LC3_HOT static void resample_48k_12k8(
337 struct lc3_ltpf_hp50_state *hp50, const int16_t *x, int16_t *y, int n)
338 {
339 resample_x192k_12k8(4, h_48k_12k8_q15, hp50, x, y, n);
340 }
341 #endif /* resample_48k_12k8 */
342
343 /**
344 * Resample to 6.4 KHz
345 * x [-3..-1] Previous, [0..n-1] Current samples
346 * y, n [0..n-1] Output `n` processed samples
347 *
348 * The `x` vector is aligned on 32 bits
349 */
350 #ifndef resample_6k4
resample_6k4(const int16_t * x,int16_t * y,int n)351 LC3_HOT static void resample_6k4(const int16_t *x, int16_t *y, int n)
352 {
353 static const int16_t h[] = { 18477, 15424, 8105 };
354 const int16_t *ye = y + n;
355
356 for (x--; y < ye; x += 2)
357 *(y++) = (x[0] * h[0] + (x[-1] + x[1]) * h[1]
358 + (x[-2] + x[2]) * h[2]) >> 16;
359 }
360 #endif /* resample_6k4 */
361
362 /**
363 * LTPF Resample to 12.8 KHz implementations for each samplerates
364 */
365
366 static void (* const resample_12k8[])
367 (struct lc3_ltpf_hp50_state *, const int16_t *, int16_t *, int ) =
368 {
369 [LC3_SRATE_8K ] = resample_8k_12k8,
370 [LC3_SRATE_16K] = resample_16k_12k8,
371 [LC3_SRATE_24K] = resample_24k_12k8,
372 [LC3_SRATE_32K] = resample_32k_12k8,
373 [LC3_SRATE_48K] = resample_48k_12k8,
374 };
375
376
377 /* ----------------------------------------------------------------------------
378 * Analysis
379 * -------------------------------------------------------------------------- */
380
381 /**
382 * Return dot product of 2 vectors
383 * a, b, n The 2 vectors of size `n` (> 0 and <= 128)
384 * return sum( a[i] * b[i] ), i = [0..n-1]
385 *
386 * The size `n` of vectors must be multiple of 16, and less or equal to 128
387 */
388 #ifndef dot
dot(const int16_t * a,const int16_t * b,int n)389 LC3_HOT static inline float dot(const int16_t *a, const int16_t *b, int n)
390 {
391 int64_t v = 0;
392
393 for (int i = 0; i < (n >> 4); i++)
394 for (int j = 0; j < 16; j++)
395 v += *(a++) * *(b++);
396
397 int32_t v32 = (v + (1 << 5)) >> 6;
398 return (float)v32;
399 }
400 #endif /* dot */
401
402 /**
403 * Return vector of correlations
404 * a, b, n The 2 vector of size `n` (> 0 and <= 128)
405 * y, nc Output the correlation vector of size `nc`
406 *
407 * The first vector `a` is aligned of 32 bits
408 * The size `n` of vectors is multiple of 16, and less or equal to 128
409 */
410 #ifndef correlate
correlate(const int16_t * a,const int16_t * b,int n,float * y,int nc)411 LC3_HOT static void correlate(
412 const int16_t *a, const int16_t *b, int n, float *y, int nc)
413 {
414 for (const float *ye = y + nc; y < ye; )
415 *(y++) = dot(a, b--, n);
416 }
417 #endif /* correlate */
418
419 /**
420 * Search the maximum value and returns its argument
421 * x, n The input vector of size `n`
422 * x_max Return the maximum value
423 * return Return the argument of the maximum
424 */
argmax(const float * x,int n,float * x_max)425 LC3_HOT static int argmax(const float *x, int n, float *x_max)
426 {
427 int arg = 0;
428
429 *x_max = x[arg = 0];
430 for (int i = 1; i < n; i++)
431 if (*x_max < x[i])
432 *x_max = x[arg = i];
433
434 return arg;
435 }
436
437 /**
438 * Search the maximum weithed value and returns its argument
439 * x, n The input vector of size `n`
440 * w_incr Increment of the weight
441 * x_max, xw_max Return the maximum not weighted value
442 * return Return the argument of the weigthed maximum
443 */
argmax_weighted(const float * x,int n,float w_incr,float * x_max)444 LC3_HOT static int argmax_weighted(
445 const float *x, int n, float w_incr, float *x_max)
446 {
447 int arg;
448
449 float xw_max = (*x_max = x[arg = 0]);
450 float w = 1 + w_incr;
451
452 for (int i = 1; i < n; i++, w += w_incr)
453 if (xw_max < x[i] * w)
454 xw_max = (*x_max = x[arg = i]) * w;
455
456 return arg;
457 }
458
459 /**
460 * Interpolate from pitch detected value (3.3.9.8)
461 * x, n [-2..-1] Previous, [0..n] Current input
462 * d The phase of interpolation (0 to 3)
463 * return The interpolated vector
464 *
465 * The size `n` of vectors must be multiple of 4
466 */
interpolate(const int16_t * x,int n,int d,int16_t * y)467 LC3_HOT static void interpolate(const int16_t *x, int n, int d, int16_t *y)
468 {
469 static const int16_t h4_q15[][4] = {
470 { 6877, 19121, 6877, 0 }, { 3506, 18025, 11000, 220 },
471 { 1300, 15048, 15048, 1300 }, { 220, 11000, 18025, 3506 } };
472
473 const int16_t *h = h4_q15[d];
474 int16_t x3 = x[-2], x2 = x[-1], x1, x0;
475
476 x1 = (*x++);
477 for (const int16_t *ye = y + n; y < ye; ) {
478 int32_t yn;
479
480 yn = (x0 = *(x++)) * h[0] + x1 * h[1] + x2 * h[2] + x3 * h[3];
481 *(y++) = yn >> 15;
482
483 yn = (x3 = *(x++)) * h[0] + x0 * h[1] + x1 * h[2] + x2 * h[3];
484 *(y++) = yn >> 15;
485
486 yn = (x2 = *(x++)) * h[0] + x3 * h[1] + x0 * h[2] + x1 * h[3];
487 *(y++) = yn >> 15;
488
489 yn = (x1 = *(x++)) * h[0] + x2 * h[1] + x3 * h[2] + x0 * h[3];
490 *(y++) = yn >> 15;
491 }
492 }
493
494 /**
495 * Interpolate autocorrelation (3.3.9.7)
496 * x [-4..-1] Previous, [0..4] Current input
497 * d The phase of interpolation (-3 to 3)
498 * return The interpolated value
499 */
interpolate_corr(const float * x,int d)500 LC3_HOT static float interpolate_corr(const float *x, int d)
501 {
502 static const float h4[][8] = {
503 { 1.53572770e-02, -4.72963246e-02, 8.35788573e-02, 8.98638285e-01,
504 8.35788573e-02, -4.72963246e-02, 1.53572770e-02, },
505 { 2.74547165e-03, 4.59833449e-03, -7.54404636e-02, 8.17488686e-01,
506 3.30182571e-01, -1.05835916e-01, 2.86823405e-02, -2.87456116e-03 },
507 { -3.00125103e-03, 2.95038503e-02, -1.30305021e-01, 6.03297008e-01,
508 6.03297008e-01, -1.30305021e-01, 2.95038503e-02, -3.00125103e-03 },
509 { -2.87456116e-03, 2.86823405e-02, -1.05835916e-01, 3.30182571e-01,
510 8.17488686e-01, -7.54404636e-02, 4.59833449e-03, 2.74547165e-03 },
511 };
512
513 const float *h = h4[(4+d) % 4];
514
515 float y = d < 0 ? x[-4] * *(h++) :
516 d > 0 ? x[ 4] * *(h+7) : 0;
517
518 y += x[-3] * h[0] + x[-2] * h[1] + x[-1] * h[2] + x[0] * h[3] +
519 x[ 1] * h[4] + x[ 2] * h[5] + x[ 3] * h[6];
520
521 return y;
522 }
523
524 /**
525 * Pitch detection algorithm (3.3.9.5-6)
526 * ltpf Context of analysis
527 * x, n [-114..-17] Previous, [0..n-1] Current 6.4KHz samples
528 * tc Return the pitch-lag estimation
529 * return True when pitch present
530 *
531 * The `x` vector is aligned on 32 bits
532 */
detect_pitch(struct lc3_ltpf_analysis * ltpf,const int16_t * x,int n,int * tc)533 static bool detect_pitch(
534 struct lc3_ltpf_analysis *ltpf, const int16_t *x, int n, int *tc)
535 {
536 float rm1, rm2;
537 float r[98];
538
539 const int r0 = 17, nr = 98;
540 int k0 = LC3_MAX( 0, ltpf->tc-4);
541 int nk = LC3_MIN(nr-1, ltpf->tc+4) - k0 + 1;
542
543 correlate(x, x - r0, n, r, nr);
544
545 int t1 = argmax_weighted(r, nr, -.5f/(nr-1), &rm1);
546 int t2 = k0 + argmax(r + k0, nk, &rm2);
547
548 const int16_t *x1 = x - (r0 + t1);
549 const int16_t *x2 = x - (r0 + t2);
550
551 float nc1 = rm1 <= 0 ? 0 :
552 rm1 / sqrtf(dot(x, x, n) * dot(x1, x1, n));
553
554 float nc2 = rm2 <= 0 ? 0 :
555 rm2 / sqrtf(dot(x, x, n) * dot(x2, x2, n));
556
557 int t1sel = nc2 <= 0.85f * nc1;
558 ltpf->tc = (t1sel ? t1 : t2);
559
560 *tc = r0 + ltpf->tc;
561 return (t1sel ? nc1 : nc2) > 0.6f;
562 }
563
564 /**
565 * Pitch-lag parameter (3.3.9.7)
566 * x, n [-232..-28] Previous, [0..n-1] Current 12.8KHz samples, Q14
567 * tc Pitch-lag estimation
568 * pitch The pitch value, in fixed .4
569 * return The bitstream pitch index value
570 *
571 * The `x` vector is aligned on 32 bits
572 */
refine_pitch(const int16_t * x,int n,int tc,int * pitch)573 static int refine_pitch(const int16_t *x, int n, int tc, int *pitch)
574 {
575 float r[17], rm;
576 int e, f;
577
578 int r0 = LC3_MAX( 32, 2*tc - 4);
579 int nr = LC3_MIN(228, 2*tc + 4) - r0 + 1;
580
581 correlate(x, x - (r0 - 4), n, r, nr + 8);
582
583 e = r0 + argmax(r + 4, nr, &rm);
584 const float *re = r + (e - (r0 - 4));
585
586 float dm = interpolate_corr(re, f = 0);
587 for (int i = 1; i <= 3; i++) {
588 float d;
589
590 if (e >= 127 && ((i & 1) | (e >= 157)))
591 continue;
592
593 if ((d = interpolate_corr(re, i)) > dm)
594 dm = d, f = i;
595
596 if (e > 32 && (d = interpolate_corr(re, -i)) > dm)
597 dm = d, f = -i;
598 }
599
600 e -= (f < 0);
601 f += 4*(f < 0);
602
603 *pitch = 4*e + f;
604 return e < 127 ? 4*e + f - 128 :
605 e < 157 ? 2*e + (f >> 1) + 126 : e + 283;
606 }
607
608 /**
609 * LTPF Analysis
610 */
lc3_ltpf_analyse(enum lc3_dt dt,enum lc3_srate sr,struct lc3_ltpf_analysis * ltpf,const int16_t * x,struct lc3_ltpf_data * data)611 bool lc3_ltpf_analyse(
612 enum lc3_dt dt, enum lc3_srate sr, struct lc3_ltpf_analysis *ltpf,
613 const int16_t *x, struct lc3_ltpf_data *data)
614 {
615 /* --- Resampling to 12.8 KHz --- */
616
617 int z_12k8 = sizeof(ltpf->x_12k8) / sizeof(*ltpf->x_12k8);
618 int n_12k8 = dt == LC3_DT_7M5 ? 96 : 128;
619
620 memmove(ltpf->x_12k8, ltpf->x_12k8 + n_12k8,
621 (z_12k8 - n_12k8) * sizeof(*ltpf->x_12k8));
622
623 int16_t *x_12k8 = ltpf->x_12k8 + (z_12k8 - n_12k8);
624
625 resample_12k8[sr](<pf->hp50, x, x_12k8, n_12k8);
626
627 x_12k8 -= (dt == LC3_DT_7M5 ? 44 : 24);
628
629 /* --- Resampling to 6.4 KHz --- */
630
631 int z_6k4 = sizeof(ltpf->x_6k4) / sizeof(*ltpf->x_6k4);
632 int n_6k4 = n_12k8 >> 1;
633
634 memmove(ltpf->x_6k4, ltpf->x_6k4 + n_6k4,
635 (z_6k4 - n_6k4) * sizeof(*ltpf->x_6k4));
636
637 int16_t *x_6k4 = ltpf->x_6k4 + (z_6k4 - n_6k4);
638
639 resample_6k4(x_12k8, x_6k4, n_6k4);
640
641 /* --- Pitch detection --- */
642
643 int tc, pitch = 0;
644 float nc = 0;
645
646 bool pitch_present = detect_pitch(ltpf, x_6k4, n_6k4, &tc);
647
648 if (pitch_present) {
649 int16_t u[128], v[128];
650
651 data->pitch_index = refine_pitch(x_12k8, n_12k8, tc, &pitch);
652
653 interpolate(x_12k8, n_12k8, 0, u);
654 interpolate(x_12k8 - (pitch >> 2), n_12k8, pitch & 3, v);
655
656 nc = dot(u, v, n_12k8) / sqrtf(dot(u, u, n_12k8) * dot(v, v, n_12k8));
657 }
658
659 /* --- Activation --- */
660
661 if (ltpf->active) {
662 int pitch_diff =
663 LC3_MAX(pitch, ltpf->pitch) - LC3_MIN(pitch, ltpf->pitch);
664 float nc_diff = nc - ltpf->nc[0];
665
666 data->active = pitch_present &&
667 ((nc > 0.9f) || (nc > 0.84f && pitch_diff < 8 && nc_diff > -0.1f));
668
669 } else {
670 data->active = pitch_present &&
671 ( (dt == LC3_DT_10M || ltpf->nc[1] > 0.94f) &&
672 (ltpf->nc[0] > 0.94f && nc > 0.94f) );
673 }
674
675 ltpf->active = data->active;
676 ltpf->pitch = pitch;
677 ltpf->nc[1] = ltpf->nc[0];
678 ltpf->nc[0] = nc;
679
680 return pitch_present;
681 }
682
683
684 /* ----------------------------------------------------------------------------
685 * Synthesis
686 * -------------------------------------------------------------------------- */
687
688 /**
689 * Width of synthesis filter
690 */
691
692 #define FILTER_WIDTH(sr) \
693 LC3_MAX(4, LC3_SRATE_KHZ(sr) / 4)
694
695 #define MAX_FILTER_WIDTH \
696 FILTER_WIDTH(LC3_NUM_SRATE)
697
698
699 /**
700 * Synthesis filter template
701 * xh, nh History ring buffer of filtered samples
702 * lag Lag parameter in the ring buffer
703 * x0 w-1 previous input samples
704 * x, n Current samples as input, filtered as output
705 * c, w Coefficients `den` then `num`, and width of filter
706 * fade Fading mode of filter -1: Out 1: In 0: None
707 */
synthesize_template(const float * xh,int nh,int lag,const float * x0,float * x,int n,const float * c,const int w,int fade)708 LC3_HOT static inline void synthesize_template(
709 const float *xh, int nh, int lag,
710 const float *x0, float *x, int n,
711 const float *c, const int w, int fade)
712 {
713 float g = (float)(fade <= 0);
714 float g_incr = (float)((fade > 0) - (fade < 0)) / n;
715 float u[MAX_FILTER_WIDTH];
716
717 /* --- Load previous samples --- */
718
719 lag += (w >> 1);
720
721 const float *y = x - xh < lag ? x + (nh - lag) : x - lag;
722 const float *y_end = xh + nh - 1;
723
724 for (int j = 0; j < w-1; j++) {
725
726 u[j] = 0;
727
728 float yi = *y, xi = *(x0++);
729 y = y < y_end ? y + 1 : xh;
730
731 for (int k = 0; k <= j; k++)
732 u[j-k] -= yi * c[k];
733
734 for (int k = 0; k <= j; k++)
735 u[j-k] += xi * c[w+k];
736 }
737
738 u[w-1] = 0;
739
740 /* --- Process by filter length --- */
741
742 for (int i = 0; i < n; i += w)
743 for (int j = 0; j < w; j++, g += g_incr) {
744
745 float yi = *y, xi = *x;
746 y = y < y_end ? y + 1 : xh;
747
748 for (int k = 0; k < w; k++)
749 u[(j+(w-1)-k)%w] -= yi * c[k];
750
751 for (int k = 0; k < w; k++)
752 u[(j+(w-1)-k)%w] += xi * c[w+k];
753
754 *(x++) = xi - g * u[j];
755 u[j] = 0;
756 }
757 }
758
759 /**
760 * Synthesis filter for each samplerates (width of filter)
761 */
762
763
synthesize_4(const float * xh,int nh,int lag,const float * x0,float * x,int n,const float * c,int fade)764 LC3_HOT static void synthesize_4(const float *xh, int nh, int lag,
765 const float *x0, float *x, int n, const float *c, int fade)
766 {
767 synthesize_template(xh, nh, lag, x0, x, n, c, 4, fade);
768 }
769
synthesize_6(const float * xh,int nh,int lag,const float * x0,float * x,int n,const float * c,int fade)770 LC3_HOT static void synthesize_6(const float *xh, int nh, int lag,
771 const float *x0, float *x, int n, const float *c, int fade)
772 {
773 synthesize_template(xh, nh, lag, x0, x, n, c, 6, fade);
774 }
775
synthesize_8(const float * xh,int nh,int lag,const float * x0,float * x,int n,const float * c,int fade)776 LC3_HOT static void synthesize_8(const float *xh, int nh, int lag,
777 const float *x0, float *x, int n, const float *c, int fade)
778 {
779 synthesize_template(xh, nh, lag, x0, x, n, c, 8, fade);
780 }
781
synthesize_12(const float * xh,int nh,int lag,const float * x0,float * x,int n,const float * c,int fade)782 LC3_HOT static void synthesize_12(const float *xh, int nh, int lag,
783 const float *x0, float *x, int n, const float *c, int fade)
784 {
785 synthesize_template(xh, nh, lag, x0, x, n, c, 12, fade);
786 }
787
788 static void (* const synthesize[])(const float *, int, int,
789 const float *, float *, int, const float *, int) =
790 {
791 [LC3_SRATE_8K ] = synthesize_4,
792 [LC3_SRATE_16K] = synthesize_4,
793 [LC3_SRATE_24K] = synthesize_6,
794 [LC3_SRATE_32K] = synthesize_8,
795 [LC3_SRATE_48K] = synthesize_12,
796 };
797
798
799 /**
800 * LTPF Synthesis
801 */
lc3_ltpf_synthesize(enum lc3_dt dt,enum lc3_srate sr,int nbytes,lc3_ltpf_synthesis_t * ltpf,const lc3_ltpf_data_t * data,const float * xh,float * x)802 void lc3_ltpf_synthesize(enum lc3_dt dt, enum lc3_srate sr, int nbytes,
803 lc3_ltpf_synthesis_t *ltpf, const lc3_ltpf_data_t *data,
804 const float *xh, float *x)
805 {
806 int nh = LC3_NH(dt, sr);
807 int dt_us = LC3_DT_US(dt);
808
809 /* --- Filter parameters --- */
810
811 int p_idx = data ? data->pitch_index : 0;
812 int pitch =
813 p_idx >= 440 ? (((p_idx ) - 283) << 2) :
814 p_idx >= 380 ? (((p_idx >> 1) - 63) << 2) + (((p_idx & 1)) << 1) :
815 (((p_idx >> 2) + 32) << 2) + (((p_idx & 3)) << 0) ;
816
817 pitch = (pitch * LC3_SRATE_KHZ(sr) * 10 + 64) / 128;
818
819 int nbits = (nbytes*8 * 10000 + (dt_us/2)) / dt_us;
820 int g_idx = LC3_MAX(nbits / 80, 3 + (int)sr) - (3 + sr);
821 bool active = data && data->active && g_idx < 4;
822
823 int w = FILTER_WIDTH(sr);
824 float c[2 * MAX_FILTER_WIDTH];
825
826 for (int i = 0; i < w; i++) {
827 float g = active ? 0.4f - 0.05f * g_idx : 0;
828 c[ i] = g * lc3_ltpf_cden[sr][pitch & 3][(w-1)-i];
829 c[w+i] = 0.85f * g * lc3_ltpf_cnum[sr][LC3_MIN(g_idx, 3)][(w-1)-i];
830 }
831
832 /* --- Transition handling --- */
833
834 int ns = LC3_NS(dt, sr);
835 int nt = ns / (3 + dt);
836 float x0[MAX_FILTER_WIDTH];
837
838 if (active)
839 memcpy(x0, x + nt-(w-1), (w-1) * sizeof(float));
840
841 if (!ltpf->active && active)
842 synthesize[sr](xh, nh, pitch/4, ltpf->x, x, nt, c, 1);
843 else if (ltpf->active && !active)
844 synthesize[sr](xh, nh, ltpf->pitch/4, ltpf->x, x, nt, ltpf->c, -1);
845 else if (ltpf->active && active && ltpf->pitch == pitch)
846 synthesize[sr](xh, nh, pitch/4, ltpf->x, x, nt, c, 0);
847 else if (ltpf->active && active) {
848 synthesize[sr](xh, nh, ltpf->pitch/4, ltpf->x, x, nt, ltpf->c, -1);
849 synthesize[sr](xh, nh, pitch/4,
850 (x <= xh ? x + nh : x) - (w-1), x, nt, c, 1);
851 }
852
853 /* --- Remainder --- */
854
855 memcpy(ltpf->x, x + ns - (w-1), (w-1) * sizeof(float));
856
857 if (active)
858 synthesize[sr](xh, nh, pitch/4, x0, x + nt, ns-nt, c, 0);
859
860 /* --- Update state --- */
861
862 ltpf->active = active;
863 ltpf->pitch = pitch;
864 memcpy(ltpf->c, c, 2*w * sizeof(*ltpf->c));
865 }
866
867
868 /* ----------------------------------------------------------------------------
869 * Bitstream data
870 * -------------------------------------------------------------------------- */
871
872 /**
873 * LTPF disable
874 */
lc3_ltpf_disable(struct lc3_ltpf_data * data)875 void lc3_ltpf_disable(struct lc3_ltpf_data *data)
876 {
877 data->active = false;
878 }
879
880 /**
881 * Return number of bits coding the bitstream data
882 */
lc3_ltpf_get_nbits(bool pitch)883 int lc3_ltpf_get_nbits(bool pitch)
884 {
885 return 1 + 10 * pitch;
886 }
887
888 /**
889 * Put bitstream data
890 */
lc3_ltpf_put_data(lc3_bits_t * bits,const struct lc3_ltpf_data * data)891 void lc3_ltpf_put_data(lc3_bits_t *bits,
892 const struct lc3_ltpf_data *data)
893 {
894 lc3_put_bit(bits, data->active);
895 lc3_put_bits(bits, data->pitch_index, 9);
896 }
897
898 /**
899 * Get bitstream data
900 */
lc3_ltpf_get_data(lc3_bits_t * bits,struct lc3_ltpf_data * data)901 void lc3_ltpf_get_data(lc3_bits_t *bits, struct lc3_ltpf_data *data)
902 {
903 data->active = lc3_get_bit(bits);
904 data->pitch_index = lc3_get_bits(bits, 9);
905 }
906
