eden-emu/opus
15
0
Fork 0
mirror of https://github.com/xiph/opus.git synced 2025-06-06 07:21:03 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 4

Fixes MSVC warnings for double->float and float->int conversion

Browse source
This commit is contained in:
Jean-Marc Valin 2012-12-21 16:17:38 -05:00
parent b5f739455d
commit d683c76b88
3 changed files with 83 additions and 84 deletions
Download patch file Download diff file Expand all files Collapse all files

40
celt/celt_encoder.c
View file

@ -215,7 +215,7 @@ static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int
int is_transient = 0; int is_transient = 0;
opus_int32 mask_metric = 0; opus_int32 mask_metric = 0;
int c; int c;
int tf_max; opus_val16 tf_max;
int len2; int len2;
/* Table of 6*64/x, trained on real data to minimize the average error */ /* Table of 6*64/x, trained on real data to minimize the average error */
static const unsigned char inv_table[128] = { static const unsigned char inv_table[128] = {
@ -318,7 +318,7 @@ static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int
/* Costs two sqrt() to avoid overflows */ /* Costs two sqrt() to avoid overflows */
mean = MULT16_16(celt_sqrt(mean), celt_sqrt(MULT16_16(maxE,len2>>1))); mean = MULT16_16(celt_sqrt(mean), celt_sqrt(MULT16_16(maxE,len2>>1)));
#else #else
mean = sqrt(mean * maxE*.5*len2); mean = celt_sqrt(mean * maxE*.5*len2);
#endif #endif
/* Inverse of the mean energy in Q15+6 */ /* Inverse of the mean energy in Q15+6 */
norm = SHL32(EXTEND32(len2),6+14)/ADD32(EPSILON,SHR32(mean,1)); norm = SHL32(EXTEND32(len2),6+14)/ADD32(EPSILON,SHR32(mean,1));
@ -331,7 +331,7 @@ static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int
#ifdef FIXED_POINT #ifdef FIXED_POINT
id = IMAX(0,IMIN(127,MULT16_32_Q15(tmp[i],norm))); /* Do not round to nearest */ id = IMAX(0,IMIN(127,MULT16_32_Q15(tmp[i],norm))); /* Do not round to nearest */
#else #else
id = IMAX(0,IMIN(127,floor(64*norm*tmp[i]))); /* Do not round to nearest */ id = IMAX(0,IMIN(127,(int)floor(64*norm*tmp[i]))); /* Do not round to nearest */
#endif #endif
unmask += inv_table[id]; unmask += inv_table[id];
} }
@ -349,7 +349,7 @@ static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int
/* Arbitrary metric for VBR boost */ /* Arbitrary metric for VBR boost */
tf_max = MAX16(0,celt_sqrt(27*mask_metric)-42); tf_max = MAX16(0,celt_sqrt(27*mask_metric)-42);
/* *tf_estimate = 1 + MIN16(1, sqrt(MAX16(0, tf_max-30))/20); */ /* *tf_estimate = 1 + MIN16(1, sqrt(MAX16(0, tf_max-30))/20); */
*tf_estimate = celt_sqrt(MAX16(0, SHL32(MULT16_16(QCONST16(0.0069,14),IMIN(163,tf_max)),14)-QCONST32(0.139,28))); *tf_estimate = celt_sqrt(MAX16(0, SHL32(MULT16_16(QCONST16(0.0069,14),MIN16(163,tf_max)),14)-QCONST32(0.139,28)));
/*printf("%d %f\n", tf_max, mask_metric);*/ /*printf("%d %f\n", tf_max, mask_metric);*/
RESTORE_STACK; RESTORE_STACK;
#ifdef FUZZING #ifdef FUZZING
@ -755,14 +755,14 @@ static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X,
#ifndef FIXED_POINT #ifndef FIXED_POINT
if (analysis->valid) if (analysis->valid)
{ {
trim -= MAX16(-QCONST16(2.f, 8), MIN16(QCONST16(2.f, 8), 2*(analysis->tonality_slope+.05))); trim -= MAX16(-QCONST16(2.f, 8), MIN16(QCONST16(2.f, 8), 2*(analysis->tonality_slope+.05f)));
} }
#endif #endif
#ifdef FIXED_POINT #ifdef FIXED_POINT
trim_index = PSHR32(trim, 8); trim_index = PSHR32(trim, 8);
#else #else
trim_index = floor(.5+trim); trim_index = (int)floor(.5f+trim);
#endif #endif
if (trim_index<0) if (trim_index<0)
trim_index = 0; trim_index = 0;
@ -807,7 +807,7 @@ static int stereo_analysis(const CELTMode *m, const celt_norm *X,
> MULT16_32_Q15(m->eBands[13]<<(LM+1), sumLR); > MULT16_32_Q15(m->eBands[13]<<(LM+1), sumLR);
} }
static int dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16 *bandLogE2, static opus_val16 dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16 *bandLogE2,
int nbEBands, int start, int end, int C, int *offsets, int lsb_depth, const opus_int16 *logN, int nbEBands, int start, int end, int C, int *offsets, int lsb_depth, const opus_int16 *logN,
int isTransient, int vbr, int constrained_vbr, const opus_int16 *eBands, int LM, int isTransient, int vbr, int constrained_vbr, const opus_int16 *eBands, int LM,
int effectiveBytes, opus_int32 *tot_boost_) int effectiveBytes, opus_int32 *tot_boost_)
@ -894,13 +894,13 @@ static int dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16 *bandL
width = C*(eBands[i+1]-eBands[i])<<LM; width = C*(eBands[i+1]-eBands[i])<<LM;
if (width<6) if (width<6)
{ {
boost = SHR32(EXTEND32(follower[i]),DB_SHIFT); boost = (int)SHR32(EXTEND32(follower[i]),DB_SHIFT);
boost_bits = boost*width<<BITRES; boost_bits = boost*width<<BITRES;
} else if (width > 48) { } else if (width > 48) {
boost = SHR32(EXTEND32(follower[i])*8,DB_SHIFT); boost = (int)SHR32(EXTEND32(follower[i])*8,DB_SHIFT);
boost_bits = (boost*width<<BITRES)/8; boost_bits = (boost*width<<BITRES)/8;
} else { } else {
boost = SHR32(EXTEND32(follower[i])*width/6,DB_SHIFT); boost = (int)SHR32(EXTEND32(follower[i])*width/6,DB_SHIFT);
boost_bits = boost*6<<BITRES; boost_bits = boost*6<<BITRES;
} }
/* For CBR and non-transient CVBR frames, limit dynalloc to 1/4 of the bits */ /* For CBR and non-transient CVBR frames, limit dynalloc to 1/4 of the bits */
@ -1501,7 +1501,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
/* effectiveRate in kb/s */ /* effectiveRate in kb/s */
effectiveRate = 2*effectiveRate/5; effectiveRate = 2*effectiveRate/5;
st->intensity = hysteresis_decision(effectiveRate, intensity_thresholds, intensity_histeresis, 21, st->intensity); st->intensity = hysteresis_decision((opus_val16)effectiveRate, intensity_thresholds, intensity_histeresis, 21, st->intensity);
st->intensity = IMIN(st->end,IMAX(st->start, st->intensity)); st->intensity = IMIN(st->end,IMAX(st->start, st->intensity));
} }
@ -1542,7 +1542,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
/*printf("%f %f %f %f %d %d ", st->analysis.activity, st->analysis.tonality, tf_estimate, st->stereo_saving, tot_boost, coded_bands);*/ /*printf("%f %f %f %f %d %d ", st->analysis.activity, st->analysis.tonality, tf_estimate, st->stereo_saving, tot_boost, coded_bands);*/
#ifndef FIXED_POINT #ifndef FIXED_POINT
if (st->analysis.valid && st->analysis.activity<.4) if (st->analysis.valid && st->analysis.activity<.4)
target -= (coded_bins<<BITRES)*1*(.4-st->analysis.activity); target -= (opus_int32)((coded_bins<<BITRES)*(.4f-st->analysis.activity));
#endif #endif
/* Stereo savings */ /* Stereo savings */
if (C==2) if (C==2)
@ -1555,25 +1555,25 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
/* Maximum fraction of the bits we can save if the signal is mono. */ /* Maximum fraction of the bits we can save if the signal is mono. */
max_frac = DIV32_16(MULT16_16(QCONST16(0.8f, 15), coded_stereo_dof), coded_bins); max_frac = DIV32_16(MULT16_16(QCONST16(0.8f, 15), coded_stereo_dof), coded_bins);
/*printf("%d %d %d ", coded_stereo_dof, coded_bins, tot_boost);*/ /*printf("%d %d %d ", coded_stereo_dof, coded_bins, tot_boost);*/
target -= MIN32(MULT16_32_Q15(max_frac,target), target -= (opus_int32)MIN32(MULT16_32_Q15(max_frac,target),
SHR16(MULT16_16(st->stereo_saving-QCONST16(0.1f,8),(coded_stereo_dof<<BITRES)),8)); SHR16(MULT16_16(st->stereo_saving-QCONST16(0.1f,8),(coded_stereo_dof<<BITRES)),8));
} }
/* Boost the rate according to dynalloc (minus the dynalloc average for calibration). */ /* Boost the rate according to dynalloc (minus the dynalloc average for calibration). */
target += tot_boost-(16<<LM); target += tot_boost-(16<<LM);
/* Apply transient boost, compensating for average boost. */ /* Apply transient boost, compensating for average boost. */
target += SHL32(MULT16_32_Q15(tf_estimate-QCONST16(0.04f,14), target),1); target += (opus_int32)SHL32(MULT16_32_Q15(tf_estimate-QCONST16(0.04f,14), target),1);
#ifndef FIXED_POINT #ifndef FIXED_POINT
/* Apply tonality boost */ /* Apply tonality boost */
if (st->analysis.valid) { if (st->analysis.valid) {
int tonal_target; opus_int32 tonal_target;
float tonal; float tonal;
/* Tonality boost (compensating for the average). */ /* Tonality boost (compensating for the average). */
tonal = MAX16(0,st->analysis.tonality-.15)-0.09; tonal = MAX16(0.f,st->analysis.tonality-.15f)-0.09f;
tonal_target = target + (coded_bins<<BITRES)*1.2f*tonal; tonal_target = target + (opus_int32)((coded_bins<<BITRES)*1.2f*tonal);
if (pitch_change) if (pitch_change)
tonal_target += (coded_bins<<BITRES)*.8; tonal_target += (opus_int32)((coded_bins<<BITRES)*.8f);
/*printf("%f %f ", st->analysis.tonality, tonal);*/ /*printf("%f %f ", st->analysis.tonality, tonal);*/
target = tonal_target; target = tonal_target;
} }
@ -1584,7 +1584,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
int bins; int bins;
bins = eBands[nbEBands-2]<<LM; bins = eBands[nbEBands-2]<<LM;
/*floor_depth = SHR32(MULT16_16((C*bins<<BITRES),celt_log2(SHL32(MAX16(1,sample_max),13))), DB_SHIFT);*/ /*floor_depth = SHR32(MULT16_16((C*bins<<BITRES),celt_log2(SHL32(MAX16(1,sample_max),13))), DB_SHIFT);*/
floor_depth = SHR32(MULT16_16((C*bins<<BITRES),maxDepth), DB_SHIFT); floor_depth = (opus_int32)SHR32(MULT16_16((C*bins<<BITRES),maxDepth), DB_SHIFT);
floor_depth = IMAX(floor_depth, target>>2); floor_depth = IMAX(floor_depth, target>>2);
target = IMIN(target, floor_depth); target = IMIN(target, floor_depth);
/*printf("%f %d\n", maxDepth, floor_depth);*/ /*printf("%f %d\n", maxDepth, floor_depth);*/
@ -1600,7 +1600,7 @@ int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm,
#endif #endif
if (st->constrained_vbr) if (st->constrained_vbr)
rate_factor = MIN16(rate_factor, QCONST16(0.67f, 15)); rate_factor = MIN16(rate_factor, QCONST16(0.67f, 15));
target = base_target + MULT16_32_Q15(rate_factor, target-base_target); target = base_target + (opus_int32)MULT16_32_Q15(rate_factor, target-base_target);
} }
/* Don't allow more than doubling the rate */ /* Don't allow more than doubling the rate */

116
src/analysis.c
View file

@ -45,22 +45,22 @@ extern const MLP net;
#endif #endif
static const float dct_table[128] = { static const float dct_table[128] = {
0.250000, 0.250000, 0.250000, 0.250000, 0.250000, 0.250000, 0.250000, 0.250000, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f,
0.250000, 0.250000, 0.250000, 0.250000, 0.250000, 0.250000, 0.250000, 0.250000, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f,
0.351851, 0.338330, 0.311806, 0.273300, 0.224292, 0.166664, 0.102631, 0.034654, 0.351851f, 0.338330f, 0.311806f, 0.273300f, 0.224292f, 0.166664f, 0.102631f, 0.034654f,
-0.034654, -0.102631, -0.166664, -0.224292, -0.273300, -0.311806, -0.338330, -0.351851, -0.034654f,-0.102631f,-0.166664f,-0.224292f,-0.273300f,-0.311806f,-0.338330f,-0.351851f,
0.346760, 0.293969, 0.196424, 0.068975, -0.068975, -0.196424, -0.293969, -0.346760, 0.346760f, 0.293969f, 0.196424f, 0.068975f,-0.068975f,-0.196424f,-0.293969f,-0.346760f,
-0.346760, -0.293969, -0.196424, -0.068975, 0.068975, 0.196424, 0.293969, 0.346760, -0.346760f,-0.293969f,-0.196424f,-0.068975f, 0.068975f, 0.196424f, 0.293969f, 0.346760f,
0.338330, 0.224292, 0.034654, -0.166664, -0.311806, -0.351851, -0.273300, -0.102631, 0.338330f, 0.224292f, 0.034654f,-0.166664f,-0.311806f,-0.351851f,-0.273300f,-0.102631f,
0.102631, 0.273300, 0.351851, 0.311806, 0.166664, -0.034654, -0.224292, -0.338330, 0.102631f, 0.273300f, 0.351851f, 0.311806f, 0.166664f,-0.034654f,-0.224292f,-0.338330f,
0.326641, 0.135299, -0.135299, -0.326641, -0.326641, -0.135299, 0.135299, 0.326641, 0.326641f, 0.135299f,-0.135299f,-0.326641f,-0.326641f,-0.135299f, 0.135299f, 0.326641f,
0.326641, 0.135299, -0.135299, -0.326641, -0.326641, -0.135299, 0.135299, 0.326641, 0.326641f, 0.135299f,-0.135299f,-0.326641f,-0.326641f,-0.135299f, 0.135299f, 0.326641f,
0.311806, 0.034654, -0.273300, -0.338330, -0.102631, 0.224292, 0.351851, 0.166664, 0.311806f, 0.034654f,-0.273300f,-0.338330f,-0.102631f, 0.224292f, 0.351851f, 0.166664f,
-0.166664, -0.351851, -0.224292, 0.102631, 0.338330, 0.273300, -0.034654, -0.311806, -0.166664f,-0.351851f,-0.224292f, 0.102631f, 0.338330f, 0.273300f,-0.034654f,-0.311806f,
0.293969, -0.068975, -0.346760, -0.196424, 0.196424, 0.346760, 0.068975, -0.293969, 0.293969f,-0.068975f,-0.346760f,-0.196424f, 0.196424f, 0.346760f, 0.068975f,-0.293969f,
-0.293969, 0.068975, 0.346760, 0.196424, -0.196424, -0.346760, -0.068975, 0.293969, -0.293969f, 0.068975f, 0.346760f, 0.196424f,-0.196424f,-0.346760f,-0.068975f, 0.293969f,
0.273300, -0.166664, -0.338330, 0.034654, 0.351851, 0.102631, -0.311806, -0.224292, 0.273300f,-0.166664f,-0.338330f, 0.034654f, 0.351851f, 0.102631f,-0.311806f,-0.224292f,
0.224292, 0.311806, -0.102631, -0.351851, -0.034654, 0.338330, 0.166664, -0.273300, 0.224292f, 0.311806f,-0.102631f,-0.351851f,-0.034654f, 0.338330f, 0.166664f,-0.273300f,
}; };
static const float analysis_window[240] = { static const float analysis_window[240] = {
@ -113,14 +113,14 @@ static const int extra_bands[NB_TOT_BANDS+1] = {
#define cA 0.43157974f #define cA 0.43157974f
#define cB 0.67848403f #define cB 0.67848403f
#define cC 0.08595542f #define cC 0.08595542f
#define cE (M_PI/2) #define cE ((float)M_PI/2)
static inline float fast_atan2f(float y, float x) { static inline float fast_atan2f(float y, float x) {
float x2, y2; float x2, y2;
/* Should avoid underflow on the values we'll get */ /* Should avoid underflow on the values we'll get */
if (ABS16(x)+ABS16(y)<1e-9) if (ABS16(x)+ABS16(y)<1e-9f)
{ {
x*=1e12; x*=1e12f;
y*=1e12; y*=1e12f;
} }
x2 = x*x; x2 = x*x;
y2 = y*y; y2 = y*y;
@ -159,7 +159,7 @@ void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info, CELTEnc
float max_frame_tonality; float max_frame_tonality;
/*float tw_sum=0;*/ /*float tw_sum=0;*/
float frame_noisiness; float frame_noisiness;
const float pi4 = M_PI*M_PI*M_PI*M_PI; const float pi4 = (float)(M_PI*M_PI*M_PI*M_PI);
float slope=0; float slope=0;
float frame_stationarity; float frame_stationarity;
float relativeE; float relativeE;
@ -213,26 +213,26 @@ void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info, CELTEnc
X2r = out[i].i+out[N-i].i; X2r = out[i].i+out[N-i].i;
X2i = out[N-i].r-out[i].r; X2i = out[N-i].r-out[i].r;
angle = (.5/M_PI)*fast_atan2f(X1i, X1r); angle = (float)(.5f/M_PI)*fast_atan2f(X1i, X1r);
d_angle = angle - A[i]; d_angle = angle - A[i];
d2_angle = d_angle - dA[i]; d2_angle = d_angle - dA[i];
angle2 = (.5/M_PI)*fast_atan2f(X2i, X2r); angle2 = (float)(.5f/M_PI)*fast_atan2f(X2i, X2r);
d_angle2 = angle2 - angle; d_angle2 = angle2 - angle;
d2_angle2 = d_angle2 - d_angle; d2_angle2 = d_angle2 - d_angle;
mod1 = d2_angle - floor(.5+d2_angle); mod1 = d2_angle - (float)floor(.5+d2_angle);
noisiness[i] = fabs(mod1); noisiness[i] = ABS16(mod1);
mod1 *= mod1; mod1 *= mod1;
mod1 *= mod1; mod1 *= mod1;
mod2 = d2_angle2 - floor(.5+d2_angle2); mod2 = d2_angle2 - (float)floor(.5+d2_angle2);
noisiness[i] += fabs(mod2); noisiness[i] += ABS16(mod2);
mod2 *= mod2; mod2 *= mod2;
mod2 *= mod2; mod2 *= mod2;
avg_mod = .25*(d2A[i]+2*mod1+mod2); avg_mod = .25f*(d2A[i]+2.f*mod1+mod2);
tonality[i] = 1./(1+40*16*pi4*avg_mod)-.015; tonality[i] = 1.f/(1.f+40.f*16.f*pi4*avg_mod)-.015f;
A[i] = angle2; A[i] = angle2;
dA[i] = d_angle2; dA[i] = d_angle2;
@ -267,30 +267,30 @@ void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info, CELTEnc
+ out[i].i*out[i].i + out[N-i].i*out[N-i].i; + out[i].i*out[i].i + out[N-i].i*out[N-i].i;
E += binE; E += binE;
tE += binE*tonality[i]; tE += binE*tonality[i];
nE += binE*2*(.5-noisiness[i]); nE += binE*2.f*(.5f-noisiness[i]);
} }
tonal->E[tonal->E_count][b] = E; tonal->E[tonal->E_count][b] = E;
frame_noisiness += nE/(1e-15+E); frame_noisiness += nE/(1e-15f+E);
frame_loudness += sqrt(E+1e-10); frame_loudness += celt_sqrt(E+1e-10f);
logE[b] = log(E+1e-10); logE[b] = (float)log(E+1e-10f);
tonal->lowE[b] = MIN32(logE[b], tonal->lowE[b]+.01); tonal->lowE[b] = MIN32(logE[b], tonal->lowE[b]+.01f);
tonal->highE[b] = MAX32(logE[b], tonal->highE[b]-.1); tonal->highE[b] = MAX32(logE[b], tonal->highE[b]-.1f);
if (tonal->highE[b] < tonal->lowE[b]+1) if (tonal->highE[b] < tonal->lowE[b]+1.f)
{ {
tonal->highE[b]+=.5; tonal->highE[b]+=.5f;
tonal->lowE[b]-=.5; tonal->lowE[b]-=.5f;
} }
relativeE += (logE[b]-tonal->lowE[b])/(EPSILON+tonal->highE[b]-tonal->lowE[b]); relativeE += (logE[b]-tonal->lowE[b])/(EPSILON+tonal->highE[b]-tonal->lowE[b]);
L1=L2=0; L1=L2=0;
for (i=0;i<NB_FRAMES;i++) for (i=0;i<NB_FRAMES;i++)
{ {
L1 += sqrt(tonal->E[i][b]); L1 += celt_sqrt(tonal->E[i][b]);
L2 += tonal->E[i][b]; L2 += tonal->E[i][b];
} }
stationarity = MIN16(0.99,L1/sqrt(EPSILON+NB_FRAMES*L2)); stationarity = MIN16(0.99f,L1/celt_sqrt(EPSILON+NB_FRAMES*L2));
stationarity *= stationarity; stationarity *= stationarity;
stationarity *= stationarity; stationarity *= stationarity;
frame_stationarity += stationarity; frame_stationarity += stationarity;
@ -307,7 +307,7 @@ void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info, CELTEnc
if (b>=NB_TBANDS-NB_TONAL_SKIP_BANDS) if (b>=NB_TBANDS-NB_TONAL_SKIP_BANDS)
frame_tonality -= band_tonality[b-NB_TBANDS+NB_TONAL_SKIP_BANDS]; frame_tonality -= band_tonality[b-NB_TBANDS+NB_TONAL_SKIP_BANDS];
#endif #endif
max_frame_tonality = MAX16(max_frame_tonality, (1+.03*(b-NB_TBANDS))*frame_tonality); max_frame_tonality = MAX16(max_frame_tonality, (1.f+.03f*(b-NB_TBANDS))*frame_tonality);
slope += band_tonality[b]*(b-8); slope += band_tonality[b]*(b-8);
/*printf("%f %f ", band_tonality[b], stationarity);*/ /*printf("%f %f ", band_tonality[b], stationarity);*/
tonal->prev_band_tonality[b] = band_tonality[b]; tonal->prev_band_tonality[b] = band_tonality[b];
@ -342,15 +342,15 @@ void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info, CELTEnc
} }
E = MAX32(E, tonal->meanE[b]); E = MAX32(E, tonal->meanE[b]);
/* 13 dB slope for spreading function */ /* 13 dB slope for spreading function */
bandwidth_mask = MAX32(.05*bandwidth_mask, E); bandwidth_mask = MAX32(.05f*bandwidth_mask, E);
/* Checks if band looks like stationary noise or if it's below a (trivial) masking curve */ /* Checks if band looks like stationary noise or if it's below a (trivial) masking curve */
if (E>.1*bandwidth_mask && E*1e10f > maxE && E > noise_floor) if (E>.1*bandwidth_mask && E*1e10f > maxE && E > noise_floor)
bandwidth = b; bandwidth = b;
} }
if (tonal->count<=2) if (tonal->count<=2)
bandwidth = 20; bandwidth = 20;
frame_loudness = 20*log10(frame_loudness); frame_loudness = 20*(float)log10(frame_loudness);
tonal->Etracker = MAX32(tonal->Etracker-.03, frame_loudness); tonal->Etracker = MAX32(tonal->Etracker-.03f, frame_loudness);
tonal->lowECount *= (1-alphaE); tonal->lowECount *= (1-alphaE);
if (frame_loudness < tonal->Etracker-30) if (frame_loudness < tonal->Etracker-30)
tonal->lowECount += alphaE; tonal->lowECount += alphaE;
@ -374,7 +374,7 @@ void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info, CELTEnc
info->activity = .5*(1+frame_noisiness-frame_stationarity); info->activity = .5*(1+frame_noisiness-frame_stationarity);
#endif #endif
frame_tonality = (max_frame_tonality/(NB_TBANDS-NB_TONAL_SKIP_BANDS)); frame_tonality = (max_frame_tonality/(NB_TBANDS-NB_TONAL_SKIP_BANDS));
frame_tonality = MAX16(frame_tonality, tonal->prev_tonality*.8); frame_tonality = MAX16(frame_tonality, tonal->prev_tonality*.8f);
tonal->prev_tonality = frame_tonality; tonal->prev_tonality = frame_tonality;
slope /= 8*8; slope /= 8*8;
@ -385,15 +385,15 @@ void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info, CELTEnc
info->tonality = frame_tonality; info->tonality = frame_tonality;
for (i=0;i<4;i++) for (i=0;i<4;i++)
features[i] = -0.12299*(BFCC[i]+tonal->mem[i+24]) + 0.49195*(tonal->mem[i]+tonal->mem[i+16]) + 0.69693*tonal->mem[i+8] - 1.4349*tonal->cmean[i]; features[i] = -0.12299f*(BFCC[i]+tonal->mem[i+24]) + 0.49195f*(tonal->mem[i]+tonal->mem[i+16]) + 0.69693f*tonal->mem[i+8] - 1.4349f*tonal->cmean[i];
for (i=0;i<4;i++) for (i=0;i<4;i++)
tonal->cmean[i] = (1-alpha)*tonal->cmean[i] + alpha*BFCC[i]; tonal->cmean[i] = (1-alpha)*tonal->cmean[i] + alpha*BFCC[i];
for (i=0;i<4;i++) for (i=0;i<4;i++)
features[4+i] = 0.63246*(BFCC[i]-tonal->mem[i+24]) + 0.31623*(tonal->mem[i]-tonal->mem[i+16]); features[4+i] = 0.63246f*(BFCC[i]-tonal->mem[i+24]) + 0.31623f*(tonal->mem[i]-tonal->mem[i+16]);
for (i=0;i<3;i++) for (i=0;i<3;i++)
features[8+i] = 0.53452*(BFCC[i]+tonal->mem[i+24]) - 0.26726*(tonal->mem[i]+tonal->mem[i+16]) -0.53452*tonal->mem[i+8]; features[8+i] = 0.53452f*(BFCC[i]+tonal->mem[i+24]) - 0.26726f*(tonal->mem[i]+tonal->mem[i+16]) -0.53452f*tonal->mem[i+8];
if (tonal->count > 5) if (tonal->count > 5)
{ {
@ -409,7 +409,7 @@ void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info, CELTEnc
tonal->mem[i] = BFCC[i]; tonal->mem[i] = BFCC[i];
} }
for (i=0;i<9;i++) for (i=0;i<9;i++)
features[11+i] = sqrt(tonal->std[i]); features[11+i] = celt_sqrt(tonal->std[i]);
features[20] = info->tonality; features[20] = info->tonality;
features[21] = info->activity; features[21] = info->activity;
features[22] = frame_stationarity; features[22] = frame_stationarity;
@ -418,9 +418,9 @@ void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info, CELTEnc
#ifndef FIXED_POINT #ifndef FIXED_POINT
mlp_process(&net, features, &frame_prob); mlp_process(&net, features, &frame_prob);
frame_prob = .5*(frame_prob+1); frame_prob = .5f*(frame_prob+1);
/* Curve fitting between the MLP probability and the actual probability */ /* Curve fitting between the MLP probability and the actual probability */
frame_prob = .01 + 1.21*frame_prob*frame_prob - .23*pow(frame_prob, 10); frame_prob = .01f + 1.21f*frame_prob*frame_prob - .23f*(float)pow(frame_prob, 10);
/*printf("%f\n", frame_prob);*/ /*printf("%f\n", frame_prob);*/
{ {
@ -428,20 +428,20 @@ void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info, CELTEnc
float p0, p1; float p0, p1;
float max_certainty; float max_certainty;
/* One transition every 3 minutes */ /* One transition every 3 minutes */
tau = .00005; tau = .00005f;
beta = .1; beta = .1f;
max_certainty = .01+1.f/(20+.5*tonal->last_transition); max_certainty = .01f+1.f/(20.f+.5f*tonal->last_transition);
p0 = (1-tonal->music_prob)*(1-tau) + tonal->music_prob *tau; p0 = (1-tonal->music_prob)*(1-tau) + tonal->music_prob *tau;
p1 = tonal->music_prob *(1-tau) + (1-tonal->music_prob)*tau; p1 = tonal->music_prob *(1-tau) + (1-tonal->music_prob)*tau;
p0 *= pow(1-frame_prob, beta); p0 *= (float)pow(1-frame_prob, beta);
p1 *= pow(frame_prob, beta); p1 *= (float)pow(frame_prob, beta);
tonal->music_prob = MAX16(max_certainty, MIN16(1-max_certainty, p1/(p0+p1))); tonal->music_prob = MAX16(max_certainty, MIN16(1-max_certainty, p1/(p0+p1)));
info->music_prob = tonal->music_prob; info->music_prob = tonal->music_prob;
/*printf("%f %f\n", frame_prob, info->music_prob);*/ /*printf("%f %f\n", frame_prob, info->music_prob);*/
} }
if (tonal->last_music != (tonal->music_prob>.5)) if (tonal->last_music != (tonal->music_prob>.5f))
tonal->last_transition=0; tonal->last_transition=0;
tonal->last_music = tonal->music_prob>.5; tonal->last_music = tonal->music_prob>.5f;
#else #else
info->music_prob = 0; info->music_prob = 0;
#endif #endif

11
src/mlp.c
View file

@ -36,7 +36,6 @@
#define MAX_NEURONS 100 #define MAX_NEURONS 100
#ifdef FIXED_POINT #ifdef FIXED_POINT
extern const opus_val16 tansig_table[501];
static inline opus_val16 tansig_approx(opus_val32 _x) /* Q19 */ static inline opus_val16 tansig_approx(opus_val32 _x) /* Q19 */
{ {
int i; int i;
@ -63,11 +62,11 @@ static inline opus_val16 tansig_approx(opus_val32 _x) /* Q19 */
} }
#else #else
/*extern const float tansig_table[501];*/ /*extern const float tansig_table[501];*/
static inline double tansig_approx(double x) static inline opus_val16 tansig_approx(opus_val16 x)
{ {
int i; int i;
double y, dy; opus_val16 y, dy;
double sign=1; opus_val16 sign=1;
if (x>=8) if (x>=8)
return 1; return 1;
if (x<=-8) if (x<=-8)
@ -77,8 +76,8 @@ static inline double tansig_approx(double x)
x=-x; x=-x;
sign=-1; sign=-1;
} }
i = floor(.5+25*x); i = (int)floor(.5f+25*x);
x -= .04*i; x -= .04f*i;
y = tansig_table[i]; y = tansig_table[i];
dy = 1-y*y; dy = 1-y*y;
y = y + x*dy*(1 - y*x); y = y + x*dy*(1 - y*x);