Renaming the SKP_ prefix to silk_
[opus.git] / silk / float / encode_frame_FLP.c
1 /***********************************************************************
2 Copyright (c) 2006-2011, Skype Limited. All rights reserved.
3 Redistribution and use in source and binary forms, with or without
4 modification, (subject to the limitations in the disclaimer below)
5 are permitted provided that the following conditions are met:
6 - Redistributions of source code must retain the above copyright notice,
7 this list of conditions and the following disclaimer.
8 - Redistributions in binary form must reproduce the above copyright
9 notice, this list of conditions and the following disclaimer in the
10 documentation and/or other materials provided with the distribution.
11 - Neither the name of Skype Limited, nor the names of specific
12 contributors, may be used to endorse or promote products derived from
13 this software without specific prior written permission.
14 NO EXPRESS OR IMPLIED LICENSES TO ANY PARTY'S PATENT RIGHTS ARE GRANTED
15 BY THIS LICENSE. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
16 CONTRIBUTORS ''AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING,
17 BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
18 FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
19 COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
20 INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21 NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
22 USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
23 ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24 (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
25 OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 ***********************************************************************/
27
28 #ifdef HAVE_CONFIG_H
29 #include "config.h"
30 #endif
31
32 #include "main_FLP.h"
33 #include "tuning_parameters.h"
34
35 /****************/
36 /* Encode frame */
37 /****************/
38 opus_int silk_encode_frame_FLP(
39     silk_encoder_state_FLP          *psEnc,             /* I/O  Encoder state FLP                       */
40     opus_int32                       *pnBytesOut,        /*   O  Number of payload bytes                 */
41     ec_enc                          *psRangeEnc         /* I/O  compressor data structure               */
42 )
43 {
44     silk_encoder_control_FLP sEncCtrl;
45     opus_int     i, ret = 0;
46     silk_float   *x_frame, *res_pitch_frame;
47     silk_float   xfw[ MAX_FRAME_LENGTH ];
48     silk_float   res_pitch[ 2 * MAX_FRAME_LENGTH + LA_PITCH_MAX ];
49
50 TIC(ENCODE_FRAME)
51
52     psEnc->sCmn.indices.Seed = psEnc->sCmn.frameCounter++ & 3;
53
54     /**************************************************************/
55     /* Setup Input Pointers, and insert frame in input buffer    */
56     /*************************************************************/
57     /* pointers aligned with start of frame to encode */
58     x_frame         = psEnc->x_buf + psEnc->sCmn.ltp_mem_length;    /* start of frame to encode */
59     res_pitch_frame = res_pitch    + psEnc->sCmn.ltp_mem_length;    /* start of pitch LPC residual frame */
60
61     /****************************/
62     /* Voice Activity Detection */
63     /****************************/
64 TIC(VAD)
65     ret = silk_VAD_GetSA_Q8( &psEnc->sCmn, psEnc->sCmn.inputBuf + 1 );
66 TOC(VAD)
67
68     /**************************************************/
69     /* Convert speech activity into VAD and DTX flags */
70     /**************************************************/
71     if( psEnc->sCmn.nFramesEncoded == 0 ) {
72         psEnc->sCmn.inDTX = psEnc->sCmn.useDTX;
73     }
74     if( psEnc->sCmn.speech_activity_Q8 < SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ) ) {
75         psEnc->sCmn.indices.signalType = TYPE_NO_VOICE_ACTIVITY;
76         psEnc->sCmn.noSpeechCounter++;
77         if( psEnc->sCmn.noSpeechCounter < NB_SPEECH_FRAMES_BEFORE_DTX ) {
78             psEnc->sCmn.inDTX = 0;
79         } else if( psEnc->sCmn.noSpeechCounter > MAX_CONSECUTIVE_DTX + NB_SPEECH_FRAMES_BEFORE_DTX ) {
80             psEnc->sCmn.noSpeechCounter = NB_SPEECH_FRAMES_BEFORE_DTX;
81             psEnc->sCmn.inDTX           = 0;
82         }
83         psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 0;
84     } else {
85         psEnc->sCmn.noSpeechCounter    = 0;
86         psEnc->sCmn.inDTX              = 0;
87         psEnc->sCmn.indices.signalType = TYPE_UNVOICED;
88         psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 1;
89     }
90
91     /***************************************/
92     /* Ensure smooth bandwidth transitions */
93     /***************************************/
94     silk_LP_variable_cutoff( &psEnc->sCmn.sLP, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length );
95
96     /*******************************************/
97     /* Copy new frame to front of input buffer */
98     /*******************************************/
99     silk_short2float_array( x_frame + LA_SHAPE_MS * psEnc->sCmn.fs_kHz, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length );
100
101     /* Add tiny signal to avoid high CPU load from denormalized floating point numbers */
102     for( i = 0; i < 8; i++ ) {
103         x_frame[ LA_SHAPE_MS * psEnc->sCmn.fs_kHz + i * ( psEnc->sCmn.frame_length >> 3 ) ] += ( 1 - ( i & 2 ) ) * 1e-6f;
104     }
105
106     /*****************************************/
107     /* Find pitch lags, initial LPC analysis */
108     /*****************************************/
109 TIC(FIND_PITCH)
110     silk_find_pitch_lags_FLP( psEnc, &sEncCtrl, res_pitch, x_frame );
111 TOC(FIND_PITCH)
112
113     /************************/
114     /* Noise shape analysis */
115     /************************/
116 TIC(NOISE_SHAPE_ANALYSIS)
117     silk_noise_shape_analysis_FLP( psEnc, &sEncCtrl, res_pitch_frame, x_frame );
118 TOC(NOISE_SHAPE_ANALYSIS)
119
120     /***************************************************/
121     /* Find linear prediction coefficients (LPC + LTP) */
122     /***************************************************/
123 TIC(FIND_PRED_COEF)
124     silk_find_pred_coefs_FLP( psEnc, &sEncCtrl, res_pitch, x_frame );
125 TOC(FIND_PRED_COEF)
126
127     /****************************************/
128     /* Process gains                        */
129     /****************************************/
130 TIC(PROCESS_GAINS)
131     silk_process_gains_FLP( psEnc, &sEncCtrl );
132 TOC(PROCESS_GAINS)
133
134     /*****************************************/
135     /* Prefiltering for noise shaper         */
136     /*****************************************/
137 TIC(PREFILTER)
138     silk_prefilter_FLP( psEnc, &sEncCtrl, xfw, x_frame );
139 TOC(PREFILTER)
140
141     /****************************************/
142     /* Low Bitrate Redundant Encoding       */
143     /****************************************/
144 TIC(LBRR)
145     silk_LBRR_encode_FLP( psEnc, &sEncCtrl, xfw );
146 TOC(LBRR)
147
148     /*****************************************/
149     /* Noise shaping quantization            */
150     /*****************************************/
151 TIC(NSQ)
152     silk_NSQ_wrapper_FLP( psEnc, &sEncCtrl, &psEnc->sCmn.indices, &psEnc->sCmn.sNSQ, psEnc->sCmn.pulses, xfw );
153 TOC(NSQ)
154
155     /* Update input buffer */
156     silk_memmove( psEnc->x_buf, &psEnc->x_buf[ psEnc->sCmn.frame_length ],
157         ( psEnc->sCmn.ltp_mem_length + LA_SHAPE_MS * psEnc->sCmn.fs_kHz ) * sizeof( silk_float ) );
158
159     /* Parameters needed for next frame */
160     psEnc->sCmn.prevLag        = sEncCtrl.pitchL[ psEnc->sCmn.nb_subfr - 1 ];
161     psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType;
162
163     /* Exit without entropy coding */
164     if( psEnc->sCmn.prefillFlag ) {
165         /* No payload */
166         *pnBytesOut = 0;
167         return ret;
168     }
169
170     /****************************************/
171     /* Encode Parameters                    */
172     /****************************************/
173 TIC(ENCODE_PARAMS)
174     silk_encode_indices( &psEnc->sCmn, psRangeEnc, psEnc->sCmn.nFramesEncoded, 0 );
175 TOC(ENCODE_PARAMS)
176
177     /****************************************/
178     /* Encode Excitation Signal             */
179     /****************************************/
180 TIC(ENCODE_PULSES)
181     silk_encode_pulses( psRangeEnc, psEnc->sCmn.indices.signalType, psEnc->sCmn.indices.quantOffsetType,
182         psEnc->sCmn.pulses, psEnc->sCmn.frame_length );
183 TOC(ENCODE_PULSES)
184
185     /****************************************/
186     /* Finalize payload                     */
187     /****************************************/
188     psEnc->sCmn.first_frame_after_reset = 0;
189     if( ++psEnc->sCmn.nFramesEncoded >= psEnc->sCmn.nFramesPerPacket ) {
190         /* Payload size */
191         *pnBytesOut = silk_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 );
192
193         /* Reset the number of frames in payload buffer */
194         psEnc->sCmn.nFramesEncoded = 0;
195     } else {
196         /* No payload this time */
197         *pnBytesOut = 0;
198     }
199 TOC(ENCODE_FRAME)
200
201 #ifdef SAVE_ALL_INTERNAL_DATA
202     /*DEBUG_STORE_DATA( xf.dat,                   pIn_HP_LP,                           psEnc->sCmn.frame_length * sizeof( opus_int16 ) );*/
203     /*DEBUG_STORE_DATA( xfw.dat,                  xfw,                                 psEnc->sCmn.frame_length * sizeof( silk_float ) );*/
204     DEBUG_STORE_DATA( pitchL.dat,               sEncCtrl.pitchL,                                 MAX_NB_SUBFR * sizeof( opus_int   ) );
205     DEBUG_STORE_DATA( pitchG_quantized.dat,     sEncCtrl.LTPCoef,            psEnc->sCmn.nb_subfr * LTP_ORDER * sizeof( silk_float ) );
206     DEBUG_STORE_DATA( LTPcorr.dat,              &psEnc->LTPCorr,                                                sizeof( silk_float ) );
207     DEBUG_STORE_DATA( gains.dat,                sEncCtrl.Gains,                          psEnc->sCmn.nb_subfr * sizeof( silk_float ) );
208     DEBUG_STORE_DATA( gains_indices.dat,        &psEnc->sCmn.indices.GainsIndices,       psEnc->sCmn.nb_subfr * sizeof( opus_int8  ) );
209     DEBUG_STORE_DATA( quantOffsetType.dat,      &psEnc->sCmn.indices.quantOffsetType,                           sizeof( opus_int8  ) );
210     DEBUG_STORE_DATA( speech_activity_q8.dat,   &psEnc->sCmn.speech_activity_Q8,                                sizeof( opus_int   ) );
211     DEBUG_STORE_DATA( signalType.dat,           &psEnc->sCmn.indices.signalType,                                sizeof( opus_int8  ) );
212     DEBUG_STORE_DATA( lag_index.dat,            &psEnc->sCmn.indices.lagIndex,                                  sizeof( opus_int16 ) );
213     DEBUG_STORE_DATA( contour_index.dat,        &psEnc->sCmn.indices.contourIndex,                              sizeof( opus_int8  ) );
214     DEBUG_STORE_DATA( per_index.dat,            &psEnc->sCmn.indices.PERIndex,                                  sizeof( opus_int8  ) );
215     DEBUG_STORE_DATA( PredCoef.dat,             &sEncCtrl.PredCoef[ 1 ],          psEnc->sCmn.predictLPCOrder * sizeof( silk_float ) );
216     DEBUG_STORE_DATA( ltp_scale_idx.dat,        &psEnc->sCmn.indices.LTP_scaleIndex,                            sizeof( opus_int8   ) );
217     /*DEBUG_STORE_DATA( xq.dat,                   psEnc->sCmn.sNSQ.xqBuf,                psEnc->sCmn.frame_length * sizeof( silk_float ) );*/
218 #endif
219     return ret;
220 }
221
222 /* Low-Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode excitation at lower bitrate  */
223 void silk_LBRR_encode_FLP(
224     silk_encoder_state_FLP          *psEnc,             /* I/O  Encoder state FLP                       */
225     silk_encoder_control_FLP        *psEncCtrl,         /* I/O  Encoder control FLP                     */
226     const silk_float                 xfw[]               /* I    Input signal                            */
227 )
228 {
229     opus_int     k;
230     opus_int32   Gains_Q16[ MAX_NB_SUBFR ];
231     silk_float   TempGains[ MAX_NB_SUBFR ];
232     SideInfoIndices *psIndices_LBRR = &psEnc->sCmn.indices_LBRR[ psEnc->sCmn.nFramesEncoded ];
233     silk_nsq_state sNSQ_LBRR;
234
235     /*******************************************/
236     /* Control use of inband LBRR              */
237     /*******************************************/
238     if( psEnc->sCmn.LBRR_enabled && psEnc->sCmn.speech_activity_Q8 > SILK_FIX_CONST( LBRR_SPEECH_ACTIVITY_THRES, 8 ) ) {
239         psEnc->sCmn.LBRR_flags[ psEnc->sCmn.nFramesEncoded ] = 1;
240
241         /* Copy noise shaping quantizer state and quantization indices from regular encoding */
242         silk_memcpy( &sNSQ_LBRR, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) );
243         silk_memcpy( psIndices_LBRR, &psEnc->sCmn.indices, sizeof( SideInfoIndices ) );
244
245         /* Save original gains */
246         silk_memcpy( TempGains, psEncCtrl->Gains, psEnc->sCmn.nb_subfr * sizeof( silk_float ) );
247
248         if( psEnc->sCmn.nFramesEncoded == 0 || psEnc->sCmn.LBRR_flags[ psEnc->sCmn.nFramesEncoded - 1 ] == 0 ) {
249             /* First frame in packet or previous frame not LBRR coded */
250             psEnc->sCmn.LBRRprevLastGainIndex = psEnc->sShape.LastGainIndex;
251
252             /* Increase Gains to get target LBRR rate */
253             psIndices_LBRR->GainsIndices[ 0 ] += psEnc->sCmn.LBRR_GainIncreases;
254             psIndices_LBRR->GainsIndices[ 0 ] = silk_min_int( psIndices_LBRR->GainsIndices[ 0 ], N_LEVELS_QGAIN - 1 );
255         }
256
257         /* Decode to get gains in sync with decoder */
258         silk_gains_dequant( Gains_Q16, psIndices_LBRR->GainsIndices,
259             &psEnc->sCmn.LBRRprevLastGainIndex, psEnc->sCmn.nFramesEncoded, psEnc->sCmn.nb_subfr );
260
261         /* Overwrite unquantized gains with quantized gains and convert back to Q0 from Q16 */
262         for( k = 0; k <  psEnc->sCmn.nb_subfr; k++ ) {
263             psEncCtrl->Gains[ k ] = Gains_Q16[ k ] * ( 1.0f / 65536.0f );
264         }
265
266         /*****************************************/
267         /* Noise shaping quantization            */
268         /*****************************************/
269         silk_NSQ_wrapper_FLP( psEnc, psEncCtrl, psIndices_LBRR, &sNSQ_LBRR,
270             psEnc->sCmn.pulses_LBRR[ psEnc->sCmn.nFramesEncoded ], xfw );
271
272         /* Restore original gains */
273         silk_memcpy( psEncCtrl->Gains, TempGains, psEnc->sCmn.nb_subfr * sizeof( silk_float ) );
274     }
275 }