3 * Copyright (c) 2004 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * @author Michael Niedermayer <michaelni@gmx.at>
34 #define FILTER_SHIFT 15
37 #define FELEM2 int32_t
38 #define FELEM_MAX INT16_MAX
39 #define FELEM_MIN INT16_MIN
41 #define FILTER_SHIFT 22
44 #define FELEM2 int64_t
45 #define FELEM_MAX INT32_MAX
46 #define FELEM_MIN INT32_MIN
50 typedef struct AVResampleContext{
58 int compensation_distance;
65 * 0th order modified bessel function of the first kind.
67 static double bessel(double x){
74 v += pow(x*x/4, i)/(t*t);
80 * builds a polyphase filterbank.
81 * @param factor resampling factor
82 * @param scale wanted sum of coefficients for each filter
83 * @param type 0->cubic, 1->blackman nuttall windowed sinc, 2->kaiser windowed sinc beta=16
85 void av_build_filter(FELEM *filter, double factor, int tap_count, int phase_count, int scale, int type){
87 double x, y, w, tab[tap_count];
88 const int center= (tap_count-1)/2;
90 /* if upsampling, only need to interpolate, no filter */
94 for(ph=0;ph<phase_count;ph++) {
97 for(i=0;i<tap_count;i++) {
98 x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor;
103 const float d= -0.5; //first order derivative = -0.5
104 x = fabs(((double)(i - center) - (double)ph / phase_count) * factor);
105 if(x<1.0) y= 1 - 3*x*x + 2*x*x*x + d*( -x*x + x*x*x);
106 else y= d*(-4 + 8*x - 5*x*x + x*x*x);
109 w = 2.0*x / (factor*tap_count) + M_PI;
110 y *= 0.3635819 - 0.4891775 * cos(w) + 0.1365995 * cos(2*w) - 0.0106411 * cos(3*w);
113 w = 2.0*x / (factor*tap_count*M_PI);
114 y *= bessel(16*sqrt(FFMAX(1-w*w, 0)));
122 /* normalize so that an uniform color remains the same */
123 for(i=0;i<tap_count;i++) {
124 v = clip(lrintf(tab[i] * scale / norm + e), FELEM_MIN, FELEM_MAX);
125 filter[ph * tap_count + i] = v;
126 e += tab[i] * scale / norm - v;
132 * initalizes a audio resampler.
133 * note, if either rate is not a integer then simply scale both rates up so they are
135 AVResampleContext *av_resample_init(int out_rate, int in_rate, int filter_size, int phase_shift, int linear, double cutoff){
136 AVResampleContext *c= av_mallocz(sizeof(AVResampleContext));
137 double factor= FFMIN(out_rate * cutoff / in_rate, 1.0);
138 int phase_count= 1<<phase_shift;
140 c->phase_shift= phase_shift;
141 c->phase_mask= phase_count-1;
144 c->filter_length= FFMAX((int)ceil(filter_size/factor), 1);
145 c->filter_bank= av_mallocz(c->filter_length*(phase_count+1)*sizeof(FELEM));
146 av_build_filter(c->filter_bank, factor, c->filter_length, phase_count, 1<<FILTER_SHIFT, 1);
147 memcpy(&c->filter_bank[c->filter_length*phase_count+1], c->filter_bank, (c->filter_length-1)*sizeof(FELEM));
148 c->filter_bank[c->filter_length*phase_count]= c->filter_bank[c->filter_length - 1];
150 c->src_incr= out_rate;
151 c->ideal_dst_incr= c->dst_incr= in_rate * phase_count;
152 c->index= -phase_count*((c->filter_length-1)/2);
157 void av_resample_close(AVResampleContext *c){
158 av_freep(&c->filter_bank);
163 * Compensates samplerate/timestamp drift. The compensation is done by changing
164 * the resampler parameters, so no audible clicks or similar distortions ocur
165 * @param compensation_distance distance in output samples over which the compensation should be performed
166 * @param sample_delta number of output samples which should be output less
168 * example: av_resample_compensate(c, 10, 500)
169 * here instead of 510 samples only 500 samples would be output
171 * note, due to rounding the actual compensation might be slightly different,
172 * especially if the compensation_distance is large and the in_rate used during init is small
174 void av_resample_compensate(AVResampleContext *c, int sample_delta, int compensation_distance){
175 // sample_delta += (c->ideal_dst_incr - c->dst_incr)*(int64_t)c->compensation_distance / c->ideal_dst_incr;
176 c->compensation_distance= compensation_distance;
177 c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr * (int64_t)sample_delta / compensation_distance;
182 * @param src an array of unconsumed samples
183 * @param consumed the number of samples of src which have been consumed are returned here
184 * @param src_size the number of unconsumed samples available
185 * @param dst_size the amount of space in samples available in dst
186 * @param update_ctx if this is 0 then the context wont be modified, that way several channels can be resampled with the same context
187 * @return the number of samples written in dst or -1 if an error occured
189 int av_resample(AVResampleContext *c, short *dst, short *src, int *consumed, int src_size, int dst_size, int update_ctx){
193 int dst_incr_frac= c->dst_incr % c->src_incr;
194 int dst_incr= c->dst_incr / c->src_incr;
195 int compensation_distance= c->compensation_distance;
197 if(compensation_distance == 0 && c->filter_length == 1 && c->phase_shift==0){
198 int64_t index2= ((int64_t)index)<<32;
199 int64_t incr= (1LL<<32) * c->dst_incr / c->src_incr;
200 dst_size= FFMIN(dst_size, (src_size-1-index) * (int64_t)c->src_incr / c->dst_incr);
202 for(dst_index=0; dst_index < dst_size; dst_index++){
203 dst[dst_index] = src[index2>>32];
206 frac += dst_index * dst_incr_frac;
207 index += dst_index * dst_incr;
208 index += frac / c->src_incr;
211 for(dst_index=0; dst_index < dst_size; dst_index++){
212 FELEM *filter= c->filter_bank + c->filter_length*(index & c->phase_mask);
213 int sample_index= index >> c->phase_shift;
216 if(sample_index < 0){
217 for(i=0; i<c->filter_length; i++)
218 val += src[FFABS(sample_index + i) % src_size] * filter[i];
219 }else if(sample_index + c->filter_length > src_size){
223 int sub_phase= (frac<<8) / c->src_incr;
224 for(i=0; i<c->filter_length; i++){
225 int64_t coeff= filter[i]*(256 - sub_phase) + filter[i + c->filter_length]*sub_phase;
226 v += src[sample_index + i] * coeff;
230 for(i=0; i<c->filter_length; i++){
231 val += src[sample_index + i] * (FELEM2)filter[i];
235 val = (val + (1<<(FILTER_SHIFT-1)))>>FILTER_SHIFT;
236 dst[dst_index] = (unsigned)(val + 32768) > 65535 ? (val>>31) ^ 32767 : val;
238 frac += dst_incr_frac;
240 if(frac >= c->src_incr){
245 if(dst_index + 1 == compensation_distance){
246 compensation_distance= 0;
247 dst_incr_frac= c->ideal_dst_incr % c->src_incr;
248 dst_incr= c->ideal_dst_incr / c->src_incr;
252 *consumed= FFMAX(index, 0) >> c->phase_shift;
253 if(index>=0) index &= c->phase_mask;
255 if(compensation_distance){
256 compensation_distance -= dst_index;
257 assert(compensation_distance > 0);
262 c->dst_incr= dst_incr_frac + c->src_incr*dst_incr;
263 c->compensation_distance= compensation_distance;
266 if(update_ctx && !c->compensation_distance){
268 av_resample_compensate(c, rand() % (8000*2) - 8000, 8000*2);
269 av_log(NULL, AV_LOG_DEBUG, "%d %d %d\n", c->dst_incr, c->ideal_dst_incr, c->compensation_distance);