curioustorvald/tsvm
1
0
Fork 0
mirror of https://github.com/curioustorvald/tsvm.git synced 2026-03-07 11:51:49 +09:00
Code Issues Packages Projects Releases Wiki Activity

TAD: even the slight companding vastly improves low-volume samples but also increases encoded size by a lot

Browse Source
This commit is contained in:
minjaesong
2025-10-27 01:22:36 +09:00
parent 67f7c091eb
commit 9c27d114fc
3 changed files with 30 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
video_encoder/decoder_tad.c
View File

@@ -277,6 +277,18 @@ static void expand_gamma(float *left, float *right, size_t count) {
}
}
static void expand_mu_law(float *left, float *right, size_t count) {
static float MU = 255.0f;
for (size_t i = 0; i < count; i++) {
// decode(y) = sign(y) * |y|^(1/γ) where γ=0.5
float x = left[i];
left[i] = signum(x) * (powf(1.0f + MU, fabsf(x)) - 1.0f) / MU;
float y = right[i];
right[i] = signum(y) * (powf(1.0f + MU, fabsf(y)) - 1.0f) / MU;
}
}
static void pcm32f_to_pcm8(const float *fleft, const float *fright, uint8_t *left, uint8_t *right, size_t count, float dither_error[2][2]) {
const float b1 = 1.5f; // 1st feedback coefficient
const float b2 = -0.75f; // 2nd feedback coefficient
@@ -321,7 +333,7 @@ static void pcm32f_to_pcm8(const float *fleft, const float *fright, uint8_t *lef
//=============================================================================
#define LAMBDA_FIXED 5.8f
#define LAMBDA_FIXED 5.0f
// Lambda-based decompanding decoder (inverse of Laplacian CDF-based encoder)
// Converts quantized index back to normalized float in [-1, 1]
@@ -523,7 +535,7 @@ static int decode_chunk(const uint8_t *input, size_t input_size, uint8_t *pcmu8_
ms_correlate(dwt_mid, dwt_side, pcm32_left, pcm32_right, sample_count);
// expand dynamic range
// expand_gamma(pcm32_left, pcm32_right, sample_count);
expand_gamma(pcm32_left, pcm32_right, sample_count);
// dither to 8-bit
pcm32f_to_pcm8(pcm32_left, pcm32_right, pcm8_left, pcm8_right, sample_count, err);

16
video_encoder/encoder_tad.c
View File

@@ -215,11 +215,23 @@ static void compress_gamma(float *left, float *right, size_t count) {
}
}
static void compress_mu_law(float *left, float *right, size_t count) {
static float MU = 255.0f;
for (size_t i = 0; i < count; i++) {
// encode(x) = sign(x) * |x|^γ where γ=0.5
float x = left[i];
left[i] = signum(x) * logf(1.0f + MU * fabsf(x)) / logf(1.0f + MU);
float y = right[i];
right[i] = signum(y) * logf(1.0f + MU * fabsf(y)) / logf(1.0f + MU);
}
}
//=============================================================================
// Quantization with Frequency-Dependent Weighting
//=============================================================================
#define LAMBDA_FIXED 5.8f
#define LAMBDA_FIXED 5.0f
// Lambda-based companding encoder (based on Laplacian distribution CDF)
// val must be normalised to [-1,1]
@@ -651,7 +663,7 @@ size_t tad32_encode_chunk(const float *pcm32_stereo, size_t num_samples,
}
// Step 1.1: Compress dynamic range
// compress_gamma(pcm32_left, pcm32_right, num_samples);
compress_gamma(pcm32_left, pcm32_right, num_samples);
// Step 2: M/S decorrelation
ms_decorrelate(pcm32_left, pcm32_right, pcm32_mid, pcm32_side, num_samples);

4
video_encoder/encoder_tad_standalone.c
View File

@@ -85,8 +85,8 @@ int main(int argc, char *argv[]) {
break;
case 'q':
quant_bits = atoi(optarg);
if (quant_bits < 4 || quant_bits > 12) {
fprintf(stderr, "Error: Quantization bits must be between 4 and 12\n");
if (quant_bits < 2 || quant_bits > 12) {
fprintf(stderr, "Error: Quantization bits must be between 2 and 12\n");
return 1;
}
break;