highperfocused/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

wmavoice: convert DCT-I/DST-I to lavu/tx

Browse Source 
This is the very last user of any lavc transform code.

This also *corrects* wmavoice decoding, as the previous DCT/DST
transforms were incorrect, bringing it closer to Microsoft's
own wmavoice decoder.
This commit is contained in:
Lynne
2023-08-04 21:16:30 +02:00
parent d895d3c8c7
commit a810126501
2 changed files with 23 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
libavcodec/wmavoice.c
View File

@@ -42,8 +42,6 @@
#include "acelp_vectors.h" #include "acelp_vectors.h"
#include "acelp_filters.h" #include "acelp_filters.h"
#include "lsp.h" #include "lsp.h"
#include "dct.h"
#include "rdft.h"
#include "sinewin.h" #include "sinewin.h"
#define MAX_BLOCKS 8 ///< maximum number of blocks per frame #define MAX_BLOCKS 8 ///< maximum number of blocks per frame
@@ -266,8 +264,8 @@ typedef struct WMAVoiceContext {
*/ */
AVTXContext *rdft, *irdft; ///< contexts for FFT-calculation in the AVTXContext *rdft, *irdft; ///< contexts for FFT-calculation in the
av_tx_fn rdft_fn, irdft_fn; ///< postfilter (for denoise filter) av_tx_fn rdft_fn, irdft_fn; ///< postfilter (for denoise filter)
DCTContext dct, dst; ///< contexts for phase shift (in Hilbert AVTXContext *dct, *dst; ///< contexts for phase shift (in Hilbert
///< transform, part of postfilter) av_tx_fn dct_fn, dst_fn; ///< transform, part of postfilter)
float sin[511], cos[511]; ///< 8-bit cosine/sine windows over [-pi,pi] float sin[511], cos[511]; ///< 8-bit cosine/sine windows over [-pi,pi]
///< range ///< range
float postfilter_agc; ///< gain control memory, used in float postfilter_agc; ///< gain control memory, used in
@@ -391,10 +389,6 @@ static av_cold int wmavoice_decode_init(AVCodecContext *ctx)
if (s->do_apf) { if (s->do_apf) {
float scale = 1.0f; float scale = 1.0f;
if ((ret = ff_dct_init (&s->dct, 6, DCT_I)) < 0 ||
(ret = ff_dct_init (&s->dst, 6, DST_I)) < 0)
return ret;
ret = av_tx_init(&s->rdft, &s->rdft_fn, AV_TX_FLOAT_RDFT, 0, 1 << 7, &scale, 0); ret = av_tx_init(&s->rdft, &s->rdft_fn, AV_TX_FLOAT_RDFT, 0, 1 << 7, &scale, 0);
if (ret < 0) if (ret < 0)
return ret; return ret;
@@ -403,6 +397,16 @@ static av_cold int wmavoice_decode_init(AVCodecContext *ctx)
if (ret < 0) if (ret < 0)
return ret; return ret;
scale = 1.0 / (1 << 6);
ret = av_tx_init(&s->dct, &s->dct_fn, AV_TX_FLOAT_DCT_I, 0, 1 << 6, &scale, 0);
if (ret < 0)
return ret;
scale = 1.0 / (1 << 6);
ret = av_tx_init(&s->dst, &s->dst_fn, AV_TX_FLOAT_DST_I, 0, 1 << 6, &scale, 0);
if (ret < 0)
return ret;
ff_sine_window_init(s->cos, 256); ff_sine_window_init(s->cos, 256);
memcpy(&s->sin[255], s->cos, 256 * sizeof(s->cos[0])); memcpy(&s->sin[255], s->cos, 256 * sizeof(s->cos[0]));
for (n = 0; n < 255; n++) { for (n = 0; n < 255; n++) {
@@ -612,6 +616,7 @@ static void calc_input_response(WMAVoiceContext *s, float *lpcs_src,
float irange, angle_mul, gain_mul, range, sq; float irange, angle_mul, gain_mul, range, sq;
LOCAL_ALIGNED_32(float, coeffs, [0x82]); LOCAL_ALIGNED_32(float, coeffs, [0x82]);
LOCAL_ALIGNED_32(float, lpcs, [0x82]); LOCAL_ALIGNED_32(float, lpcs, [0x82]);
LOCAL_ALIGNED_32(float, lpcs_dct, [0x82]);
int n, idx; int n, idx;
memcpy(coeffs, coeffs_dst, 0x82*sizeof(float)); memcpy(coeffs, coeffs_dst, 0x82*sizeof(float));
@@ -662,8 +667,8 @@ static void calc_input_response(WMAVoiceContext *s, float *lpcs_src,
* is a sine input) by doing a phase shift (in theory, H(sin())=cos()). * is a sine input) by doing a phase shift (in theory, H(sin())=cos()).
* Hilbert_Transform(RDFT(x)) = Laplace_Transform(x), which calculates the * Hilbert_Transform(RDFT(x)) = Laplace_Transform(x), which calculates the
* "moment" of the LPCs in this filter. */ * "moment" of the LPCs in this filter. */
s->dct.dct_calc(&s->dct, lpcs); s->dct_fn(s->dct, lpcs_dct, lpcs, sizeof(float));
s->dst.dct_calc(&s->dst, lpcs); s->dst_fn(s->dst, lpcs, lpcs_dct, sizeof(float));
/* Split out the coefficient indexes into phase/magnitude pairs */ /* Split out the coefficient indexes into phase/magnitude pairs */
idx = 255 + av_clip(lpcs[64], -255, 255); idx = 255 + av_clip(lpcs[64], -255, 255);
@@ -2003,8 +2008,8 @@ static av_cold int wmavoice_decode_end(AVCodecContext *ctx)
if (s->do_apf) { if (s->do_apf) {
av_tx_uninit(&s->rdft); av_tx_uninit(&s->rdft);
av_tx_uninit(&s->irdft); av_tx_uninit(&s->irdft);
ff_dct_end(&s->dct); av_tx_uninit(&s->dct);
ff_dct_end(&s->dst); av_tx_uninit(&s->dst);
} }
return 0; return 0;

9
tests/fate/wma.mak
View File

@@ -20,17 +20,20 @@ fate-wmapro: $(FATE_WMAPRO-yes)
FATE_WMAVOICE-$(call DEMDEC, ASF, WMAVOICE) += fate-wmavoice-7k FATE_WMAVOICE-$(call DEMDEC, ASF, WMAVOICE) += fate-wmavoice-7k
fate-wmavoice-7k: CMD = pcm -i $(TARGET_SAMPLES)/wmavoice/streaming_CBR-7K.wma fate-wmavoice-7k: CMD = pcm -i $(TARGET_SAMPLES)/wmavoice/streaming_CBR-7K.wma
fate-wmavoice-7k: REF = $(SAMPLES)/wmavoice/streaming_CBR-7K.pcm fate-wmavoice-7k: REF = $(SAMPLES)/wmavoice/streaming_CBR-7K_ref.pcm
fate-wmavoice-7k: CMP_TARGET = 1368.61
fate-wmavoice-7k: FUZZ = 3 fate-wmavoice-7k: FUZZ = 3
FATE_WMAVOICE-$(call DEMDEC, ASF, WMAVOICE) += fate-wmavoice-11k FATE_WMAVOICE-$(call DEMDEC, ASF, WMAVOICE) += fate-wmavoice-11k
fate-wmavoice-11k: CMD = pcm -i $(TARGET_SAMPLES)/wmavoice/streaming_CBR-11K.wma fate-wmavoice-11k: CMD = pcm -i $(TARGET_SAMPLES)/wmavoice/streaming_CBR-11K.wma
fate-wmavoice-11k: REF = $(SAMPLES)/wmavoice/streaming_CBR-11K.pcm fate-wmavoice-11k: REF = $(SAMPLES)/wmavoice/streaming_CBR-11K_ref.pcm
fate-wmavoice-11k: CMP_TARGET = 965.24
fate-wmavoice-11k: FUZZ = 3 fate-wmavoice-11k: FUZZ = 3
FATE_WMAVOICE-$(call DEMDEC, ASF, WMAVOICE) += fate-wmavoice-19k FATE_WMAVOICE-$(call DEMDEC, ASF, WMAVOICE) += fate-wmavoice-19k
fate-wmavoice-19k: CMD = pcm -i $(TARGET_SAMPLES)/wmavoice/streaming_CBR-19K.wma fate-wmavoice-19k: CMD = pcm -i $(TARGET_SAMPLES)/wmavoice/streaming_CBR-19K.wma
fate-wmavoice-19k: REF = $(SAMPLES)/wmavoice/streaming_CBR-19K.pcm fate-wmavoice-19k: REF = $(SAMPLES)/wmavoice/streaming_CBR-19K_ref.pcm
fate-wmavoice-19k: CMP_TARGET = 689.33
fate-wmavoice-19k: FUZZ = 3 fate-wmavoice-19k: FUZZ = 3
$(FATE_WMAVOICE-yes): CMP = stddev $(FATE_WMAVOICE-yes): CMP = stddev