Improved libsamplerate to allow size calculations

master
Andreas Eversberg 5 months ago
parent 1d77952717
commit 4fc92eba45
  1. 97
      src/libsamplerate/samplerate.c
  2. 1
      src/libsamplerate/samplerate.h

@ -94,15 +94,60 @@ int samplerate_downsample(samplerate_t *state, sample_t *samples, int input_num)
return output_num;
}
int samplerate_upsample_input_num(samplerate_t *state, int output_num)
{
double factor = 1.0 / state->factor, in_index;
int idx = 0;
/* count output */
in_index = state->up.in_index;
while (output_num--) {
/* increment input index */
in_index += factor;
/* count index on overflow */
if (in_index >= 1.0) {
idx++;
in_index -= 1.0;
}
}
return idx;
}
int samplerate_upsample_output_num(samplerate_t *state, int input_num)
{
double factor = 1.0 / state->factor, in_index;
int output_num = 0, idx = 0;
/* count output */
in_index = state->up.in_index;
while (idx < input_num) {
/* count output number */
output_num++;
/* increment input index */
in_index += factor;
/* count index on overflow */
if (in_index >= 1.0) {
idx++;
in_index -= 1.0;
}
}
return output_num;
}
/* convert low sample rate to high sample rate */
int samplerate_upsample(samplerate_t *state, sample_t *input, int input_num, sample_t *output)
void samplerate_upsample(samplerate_t *state, sample_t *input, int input_num, sample_t *output, int output_num)
{
int output_num = 0, i, idx;
double factor = 1.0 / state->factor, in_index, diff;
sample_t buff[(int)((double)input_num / factor + 0.5) + 10]; /* add some safety */
sample_t *samples, last_sample;
int i, idx;
double factor = 1.0 / state->factor, in_index;
sample_t buff[output_num];
sample_t *samples, current_sample, last_sample;
/* get last sample for interpolation */
current_sample = state->up.current_sample;
last_sample = state->up.last_sample;
if (input == output)
@ -112,35 +157,31 @@ int samplerate_upsample(samplerate_t *state, sample_t *input, int input_num, sam
/* resample input */
in_index = state->up.in_index;
idx = 0;
for (i = 0; ; i++) {
/* convert index to int */
idx = (int)in_index;
/* if index is outside input sample range, we are done */
if (idx >= input_num)
break;
for (i = 0; i < output_num; i++) {
/* linear interpolation */
diff = in_index - (double)idx;
if (idx)
samples[i] = input[idx - 1] * (1.0 - diff) + input[idx] * diff;
else
samples[i] = last_sample * (1.0 - diff) + input[idx] * diff;
/* count output number */
output_num++;
samples[i] = last_sample * (1.0 - in_index) + current_sample * in_index;
/* increment input index */
in_index += factor;
/* get next sample on overflow */
if (in_index >= 1.0) {
if (idx == input_num) {
fprintf(stderr, "Given input_num is too small, please fix!\n");
}
last_sample = current_sample;
current_sample = input[idx++];
in_index -= 1.0;
}
}
if (idx < input_num) {
fprintf(stderr, "Given input_num is too large, please fix!\n");
abort();
}
/* store last sample for interpolation */
if (input_num)
state->up.last_sample = input[input_num - 1];
/* remove number of input samples from index */
in_index -= (double)input_num;
/* in_index cannot be negative, except due to rounding error, so... */
if ((int)in_index < 0)
in_index = 0.0;
state->up.last_sample = last_sample;
state->up.current_sample = current_sample;
state->up.in_index = in_index;
/* filter up */
@ -151,7 +192,5 @@ int samplerate_upsample(samplerate_t *state, sample_t *input, int input_num, sam
for (i = 0; i < output_num; i++)
*output++ = samples[i];
}
return output_num;
}

@ -9,6 +9,7 @@ typedef struct samplerate {
} down;
struct {
iir_filter_t lp;
sample_t current_sample;
sample_t last_sample;
double in_index;
} up;

Loading…
Cancel
Save