Newer
Older
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
}
mvr2r( synthesis, synthesis + kk, 2 * L - kk );
mvr2r( synthesis + L, gapsynth, L );
mvr2r( prevsynth + alignment_point - kk, synthesis, kk );
return;
}
/*--------------------------------------------------------------------------
* fec_alg()
*
* Pitch based error-concealment algorithm with adaptive analysis frame
* length
*--------------------------------------------------------------------------*/
static void fec_alg(
const float *prevsynth, /* i : previous synthesis */
const float *prevsynth_LP, /* i : down-sampled synthesis */
float *ecu_rec, /* o : ECU frame with Windowing/TDA */
const int16_t output_frame, /* i : Output frame length */
const int16_t N, /* i : Analysis frame length in 8 kHz (corr. max) */
const int16_t decimatefactor, /* i : Downsampling factor */
const int16_t HqVoicing, /* i : HQ voicing flag */
float *gapsynth, /* o : Buffer for crossfade to good frame */
int16_t *ni_seed_forfec, /* i/o: Random seed for picking noise segments */
const int16_t element_mode, /* i : IVAS element mode */
const float *old_out
)
{
int16_t n, Nfft;
float sum_Tf_abs;
float Tfr[FEC_FFT_MAX_SIZE];
float Tfi[FEC_FFT_MAX_SIZE];
float Tf_abs[FEC_FFT_MAX_SIZE / 2];
float synthesis[2 * L_FRAME48k];
fec_ecu_dft( prevsynth_LP, N, Tfr, Tfi, &sum_Tf_abs, Tf_abs, &Nfft, element_mode );
sinusoidal_synthesis( Tfr, Tfi, Tf_abs, N, output_frame, decimatefactor, Nfft, sum_Tf_abs, synthesis, HqVoicing );
fec_noise_filling( prevsynth, synthesis, output_frame, N * decimatefactor, HqVoicing, gapsynth, ni_seed_forfec, element_mode, old_out );
n = (int16_t) ( (float) output_frame * N_ZERO_MDCT_NS / FRAME_SIZE_NS );
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
wtda( synthesis + ( output_frame - n ), ecu_rec, NULL, ALDO_WINDOW, ALDO_WINDOW, output_frame );
return;
}
/*--------------------------------------------------------------------------
* hq_phase_ecu()
*
* Main routine for HQ phase ECU
*--------------------------------------------------------------------------*/
static void hq_phase_ecu(
const float *prevsynth, /* i : buffer of previously synthesized signal */
float *ecu_rec, /* o : reconstructed frame in tda domain */
int16_t *time_offs, /* i/o: Sample offset for consecutive frame losses*/
float *X_sav, /* i/o: Stored spectrum of prototype frame */
int16_t *num_p, /* i/o: Number of identified peaks */
int16_t *plocs, /* i/o: Peak locations */
float *plocsi, /* i/o: Interpolated peak locations */
const float env_stab, /* i : Envelope stability parameter */
int16_t *last_fec, /* i/o: Flag for usage of pitch dependent ECU */
const int16_t prev_bfi, /* i : indicating burst frame error */
const int16_t old_is_transient[2], /* i : flags indicating previous transient frames*/
float *mag_chg_1st, /* i/o: per band magnitude modifier for transients*/
float Xavg[LGW_MAX], /* i/o: Frequency group average gain to fade to */
float *beta_mute, /* o : Factor for long-term mute */
const int16_t bwidth, /* i : Encoded bandwidth */
const int16_t output_frame, /* i : frame length */
const float pcorr, /* i : pitch correlation */
const int16_t element_mode /* i : IVAS element mode */
)
{
int16_t Lprot;
float mag_chg[LGW_MAX], ph_dith, X[L_PROT48k];
int16_t seed;
float alpha[LGW_MAX], beta[LGW_MAX];
const float *old_dec;
float noise_fac;
int16_t ph_ecu_lookahead;
noise_fac = 1.0f;
if ( element_mode == EVS_MONO )
{
ph_ecu_lookahead = NS2SA( output_frame * FRAMES_PER_SEC, PH_ECU_LOOKAHEAD_NS );
}
else
{
ph_ecu_lookahead = 0;
}
Lprot = ( 2 * output_frame * 4 ) / 5;
if ( !prev_bfi || ( prev_bfi && *last_fec && ( *time_offs == output_frame ) ) )
{
if ( !( prev_bfi && *last_fec && ( element_mode == EVS_MONO ) ) )
{
*time_offs = 0; /* IVAS reset of offset time counter, timeoffset variable later also used to calculate burst length */
}
trans_ana( prevsynth + 2 * output_frame - Lprot - *time_offs + ph_ecu_lookahead, mag_chg, &ph_dith, mag_chg_1st, output_frame, *time_offs, env_stab, *last_fec, element_mode, alpha, beta, beta_mute, Xavg );
spec_ana( prevsynth + 2 * output_frame - Lprot - *time_offs + ph_ecu_lookahead, plocs, plocsi, num_p, X_sav, output_frame, bwidth, element_mode, &noise_fac, pcorr );
if ( prev_bfi && *last_fec )
{
if ( element_mode != EVS_MONO )
Jonas Svedberg
committed
*time_offs = (int16_t) ( *time_offs + output_frame ); /* USAN avoid risk of internal int32_t in "+=" */
if ( *time_offs <= 0 )
{ /* detected wrap around of st->time_offs */
*time_offs = (int16_t) INT16_MAX; /* keep a very high value so that the long term muting stays on */
}
}
else
{
*time_offs = (int16_t) ( *time_offs + output_frame ); /* EVS_MONO BE compatible, but EVS CR needed as wrap will cause burst length muting envelope instability issues */
*time_offs = (int16_t) ( *time_offs + output_frame ); /* cast added for USAN, "+=" avoided as it may creat a truncation from int to int16_t */
if ( *time_offs <= 0 )
{
/* detect wrap around of st->time_offs */
*time_offs = (int16_t) INT16_MAX; /* high value --> continued muting will ensure that the now saturated seed is not creating tones */
trans_ana( prevsynth + 2 * output_frame - Lprot, mag_chg, &ph_dith, mag_chg_1st, output_frame, *time_offs, env_stab, 0, element_mode, alpha, beta, beta_mute, Xavg ); /* 1.0 stable-music, 0.0 speech-like */
}
mvr2r( X_sav, X, Lprot );
/* seed for own_rand2 */
seed = *time_offs;
if ( *num_p > 0 )
{
seed = (int16_t) ( seed + plocs[*num_p - 1] ); /* explicit cast and "+=" not used, as it "+=" may create a cast from 32 bit to 16 bit, triggering USAN */
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
}
subst_spec( plocs, plocsi, num_p, *time_offs, X, mag_chg, ph_dith, old_is_transient, output_frame, &seed, alpha, beta, *beta_mute, Xavg, element_mode, ph_ecu_lookahead, noise_fac );
/* reconstructed frame in tda domain */
old_dec = prevsynth + 2 * output_frame - NS2SA( output_frame * FRAMES_PER_SEC, N_ZERO_MDCT_NS );
rec_frame( X, ecu_rec, output_frame, old_dec, element_mode, num_p, plocs );
return;
}
/*--------------------------------------------------------------------------
* hq_ecu()
*
* Main routine for HQ ECU
*--------------------------------------------------------------------------*/
void hq_ecu(
const float *prevsynth, /* i : buffer of previously synthesized signal */
float *ecu_rec, /* o : reconstructed frame in tda domain */
int16_t *time_offs, /* i/o: Sample offset for consecutive frame losses*/
float *X_sav, /* i/o: Stored spectrum of prototype frame */
int16_t *num_p, /* i/o: Number of identified peaks */
int16_t *plocs, /* i/o: Peak locations */
float *plocsi, /* i/o: Interpolated peak locations */
const float env_stab, /* i : Envelope stability parameter */
int16_t *last_fec, /* i/o: Flag for usage of pitch dependent ECU */
const int16_t ph_ecu_HqVoicing, /* i : HQ Voicing flag */
int16_t *ph_ecu_active, /* i : Phase ECU active flag */
float *gapsynth, /* o : Gap synthesis for crossfade to good frame */
const int16_t prev_bfi, /* i : indicating burst frame error */
const int16_t old_is_transient[2], /* i : flags indicating previous transient frames*/
float *mag_chg_1st, /* i/o: per band magnitude modifier for transients*/
float Xavg[LGW_MAX], /* i/o: Frequency group average gain to fade to */
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
const int16_t output_frame, /* i : frame length */
Decoder_State *st /* i/o: decoder state structure */
)
{
int16_t N;
float corr = 0.0f;
int16_t decimatefactor;
float prevsynth_LP[2 * L_FRAME8k];
HQ_DEC_HANDLE hHQ_core;
const float *fec_alg_input;
int16_t evs_mode_selection;
int16_t ivas_mode_selection;
hHQ_core = st->hHQ_core;
if ( st->element_mode == EVS_MONO )
{
fec_alg_input = prevsynth + NS2SA( output_frame * FRAMES_PER_SEC, ACELP_LOOK_NS / 2 - PH_ECU_LOOKAHEAD_NS );
}
else
{
fec_alg_input = prevsynth - NS2SA( output_frame * FRAMES_PER_SEC, PH_ECU_LOOKAHEAD_NS );
}
/* find pitch and R value */
if ( !( output_frame < L_FRAME16k ) )
{
fec_ecu_pitch( fec_alg_input, prevsynth_LP, output_frame, &N, &corr, &decimatefactor, ph_ecu_HqVoicing );
}
else
{
corr = 0.0f;
decimatefactor = 4;
N = output_frame / 4;
}
evs_mode_selection = ( st->total_brate >= 48000 && ( output_frame >= L_FRAME16k && !prev_bfi && ( !old_is_transient[0] || old_is_transient[1] ) &&
( ph_ecu_HqVoicing || ( ( ( hHQ_core->env_stab_plc > 0.5 ) && ( corr < 0.6 ) ) || ( hHQ_core->env_stab_plc < 0.5 && ( corr > 0.85 ) ) ) ) ) ) ||
( st->total_brate < 48000 && ( ( ph_ecu_HqVoicing || corr > 0.85 ) && !prev_bfi && ( !old_is_transient[0] || old_is_transient[1] ) ) );
ivas_mode_selection = ( N < PH_ECU_N_LIMIT ) || ( corr < PH_ECU_CORR_LIMIT );
if ( ( ( st->element_mode == EVS_MONO ) && evs_mode_selection ) ||
( ( st->element_mode != EVS_MONO ) && evs_mode_selection && ivas_mode_selection ) )
{
fec_alg( fec_alg_input, prevsynth_LP, ecu_rec, output_frame, N, decimatefactor, ph_ecu_HqVoicing, gapsynth, &hHQ_core->ni_seed_forfec, st->element_mode, st->hHQ_core->old_out );
*last_fec = 1;
*ph_ecu_active = 0;
*time_offs = output_frame;
}
else
{
hq_phase_ecu( prevsynth - NS2SA( output_frame * FRAMES_PER_SEC, PH_ECU_LOOKAHEAD_NS ), ecu_rec, time_offs, X_sav, num_p, plocs, plocsi, env_stab, last_fec, prev_bfi, old_is_transient, mag_chg_1st, Xavg, beta_mute, st->bwidth, output_frame, corr, st->element_mode );
*last_fec = 0;
*ph_ecu_active = 1;
}
return;
}