Optimization of the CBR loop
[opus.git] / silk / dec_API.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 #include "API.h"
32 #include "main.h"
33
34 /************************/
35 /* Decoder Super Struct */
36 /************************/
37 typedef struct {
38     silk_decoder_state          channel_state[ DECODER_NUM_CHANNELS ];
39     stereo_dec_state                sStereo;
40     opus_int                         nChannelsAPI;
41     opus_int                         nChannelsInternal;
42     opus_int                         prev_decode_only_middle;
43 } silk_decoder;
44
45 /*********************/
46 /* Decoder functions */
47 /*********************/
48
49 opus_int silk_Get_Decoder_Size( int *decSizeBytes )
50 {
51     opus_int ret = SILK_NO_ERROR;
52
53     *decSizeBytes = sizeof( silk_decoder );
54
55     return ret;
56 }
57
58 /* Reset decoder state */
59 opus_int silk_InitDecoder(
60     void* decState                                      /* I/O: State                                          */
61 )
62 {
63     opus_int n, ret = SILK_NO_ERROR;
64     silk_decoder_state *channel_state = ((silk_decoder *)decState)->channel_state;
65
66     for( n = 0; n < DECODER_NUM_CHANNELS; n++ ) {
67         ret  = silk_init_decoder( &channel_state[ n ] );
68     }
69
70     return ret;
71 }
72
73 /* Decode a frame */
74 opus_int silk_Decode(
75     void*                               decState,       /* I/O: State                                           */
76     silk_DecControlStruct*      decControl,     /* I/O: Control Structure                               */
77     opus_int                             lostFlag,       /* I:   0: no loss, 1 loss, 2 decode FEC                */
78     opus_int                             newPacketFlag,  /* I:   Indicates first decoder call for this packet    */
79     ec_dec                              *psRangeDec,    /* I/O  Compressor data structure                       */
80     opus_int16                           *samplesOut,    /* O:   Decoded output speech vector                    */
81     opus_int32                           *nSamplesOut    /* O:   Number of samples decoded                       */
82 )
83 {
84     opus_int   i, n, delay, decode_only_middle = 0, ret = SILK_NO_ERROR;
85     opus_int32 nSamplesOutDec, LBRR_symbol;
86     opus_int16 samplesOut1_tmp[ 2 ][ MAX_FS_KHZ * MAX_FRAME_LENGTH_MS + 2 + MAX_DECODER_DELAY ];
87     opus_int16 samplesOut2_tmp[ MAX_API_FS_KHZ * MAX_FRAME_LENGTH_MS ];
88     opus_int32 MS_pred_Q13[ 2 ] = { 0 };
89     opus_int16 *resample_out_ptr;
90     silk_decoder *psDec = ( silk_decoder * )decState;
91     silk_decoder_state *channel_state = psDec->channel_state;
92
93     /**********************************/
94     /* Test if first frame in payload */
95     /**********************************/
96     if( newPacketFlag ) {
97         for( n = 0; n < decControl->nChannelsInternal; n++ ) {
98             channel_state[ n ].nFramesDecoded = 0;  /* Used to count frames in packet */
99         }
100     }
101
102     /* If Mono -> Stereo transition in bitstream: init state of second channel */
103     if( decControl->nChannelsInternal > psDec->nChannelsInternal ) {
104         ret += silk_init_decoder( &channel_state[ 1 ] );
105     }
106
107     if( channel_state[ 0 ].nFramesDecoded == 0 ) {
108         for( n = 0; n < decControl->nChannelsInternal; n++ ) {
109             opus_int fs_kHz_dec;
110             if( decControl->payloadSize_ms == 0 ) {
111                 /* Assuming packet loss, use 10 ms */
112                 channel_state[ n ].nFramesPerPacket = 1;
113                 channel_state[ n ].nb_subfr = 2;
114             } else if( decControl->payloadSize_ms == 10 ) {
115                 channel_state[ n ].nFramesPerPacket = 1;
116                 channel_state[ n ].nb_subfr = 2;
117             } else if( decControl->payloadSize_ms == 20 ) {
118                 channel_state[ n ].nFramesPerPacket = 1;
119                 channel_state[ n ].nb_subfr = 4;
120             } else if( decControl->payloadSize_ms == 40 ) {
121                 channel_state[ n ].nFramesPerPacket = 2;
122                 channel_state[ n ].nb_subfr = 4;
123             } else if( decControl->payloadSize_ms == 60 ) {
124                 channel_state[ n ].nFramesPerPacket = 3;
125                 channel_state[ n ].nb_subfr = 4;
126             } else {
127                 silk_assert( 0 );
128                 return SILK_DEC_INVALID_FRAME_SIZE;
129             }
130             fs_kHz_dec = ( decControl->internalSampleRate >> 10 ) + 1;
131             if( fs_kHz_dec != 8 && fs_kHz_dec != 12 && fs_kHz_dec != 16 ) {
132                 silk_assert( 0 );
133                 return SILK_DEC_INVALID_SAMPLING_FREQUENCY;
134             }
135             ret += silk_decoder_set_fs( &channel_state[ n ], fs_kHz_dec, decControl->API_sampleRate );
136         }
137     }
138
139     delay = channel_state[ 0 ].delay;
140
141     if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 2 && ( psDec->nChannelsAPI == 1 || psDec->nChannelsInternal == 1 ) ) {
142         silk_memset( psDec->sStereo.pred_prev_Q13, 0, sizeof( psDec->sStereo.pred_prev_Q13 ) );
143         silk_memset( psDec->sStereo.sSide, 0, sizeof( psDec->sStereo.sSide ) );
144         silk_memcpy( &channel_state[ 1 ].resampler_state, &channel_state[ 0 ].resampler_state, sizeof( silk_resampler_state_struct ) );
145         silk_memcpy( &channel_state[ 1 ].delayBuf, &channel_state[ 0 ].delayBuf, sizeof(channel_state[ 0 ].delayBuf));
146     }
147     psDec->nChannelsAPI      = decControl->nChannelsAPI;
148     psDec->nChannelsInternal = decControl->nChannelsInternal;
149
150     if( decControl->API_sampleRate > MAX_API_FS_KHZ * 1000 || decControl->API_sampleRate < 8000 ) {
151         ret = SILK_DEC_INVALID_SAMPLING_FREQUENCY;
152         return( ret );
153     }
154
155     if( lostFlag != FLAG_PACKET_LOST && channel_state[ 0 ].nFramesDecoded == 0 ) {
156         /* First decoder call for this payload */
157         /* Decode VAD flags and LBRR flag */
158         for( n = 0; n < decControl->nChannelsInternal; n++ ) {
159             for( i = 0; i < channel_state[ n ].nFramesPerPacket; i++ ) {
160                 channel_state[ n ].VAD_flags[ i ] = ec_dec_bit_logp(psRangeDec, 1);
161             }
162             channel_state[ n ].LBRR_flag = ec_dec_bit_logp(psRangeDec, 1);
163         }
164         /* Decode LBRR flags */
165         for( n = 0; n < decControl->nChannelsInternal; n++ ) {
166             silk_memset( channel_state[ n ].LBRR_flags, 0, sizeof( channel_state[ n ].LBRR_flags ) );
167             if( channel_state[ n ].LBRR_flag ) {
168                 if( channel_state[ n ].nFramesPerPacket == 1 ) {
169                     channel_state[ n ].LBRR_flags[ 0 ] = 1;
170                 } else {
171                     LBRR_symbol = ec_dec_icdf( psRangeDec, silk_LBRR_flags_iCDF_ptr[ channel_state[ n ].nFramesPerPacket - 2 ], 8 ) + 1;
172                     for( i = 0; i < channel_state[ n ].nFramesPerPacket; i++ ) {
173                         channel_state[ n ].LBRR_flags[ i ] = silk_RSHIFT( LBRR_symbol, i ) & 1;
174                     }
175                 }
176             }
177         }
178
179         if( lostFlag == FLAG_DECODE_NORMAL ) {
180             /* Regular decoding: skip all LBRR data */
181             for( i = 0; i < channel_state[ 0 ].nFramesPerPacket; i++ ) {
182                 for( n = 0; n < decControl->nChannelsInternal; n++ ) {
183                     if( channel_state[ n ].LBRR_flags[ i ] ) {
184                         opus_int pulses[ MAX_FRAME_LENGTH ];
185                         opus_int condCoding;
186
187                         if( decControl->nChannelsInternal == 2 && n == 0 ) {
188                             silk_stereo_decode_pred( psRangeDec, MS_pred_Q13 );
189                             if( channel_state[ 1 ].LBRR_flags[ i ] == 0 ) {
190                                 silk_stereo_decode_mid_only( psRangeDec, &decode_only_middle );
191                             }
192                         }
193                         /* Use conditional coding if previous frame available */
194                         if( i > 0 && channel_state[ n ].LBRR_flags[ i - 1 ] ) {
195                             condCoding = CODE_CONDITIONALLY;
196                         } else {
197                             condCoding = CODE_INDEPENDENTLY;
198                         }
199                         silk_decode_indices( &channel_state[ n ], psRangeDec, i, 1, condCoding );
200                         silk_decode_pulses( psRangeDec, pulses, channel_state[ n ].indices.signalType,
201                             channel_state[ n ].indices.quantOffsetType, channel_state[ n ].frame_length );
202                     }
203                 }
204             }
205         }
206     }
207
208     /* Get MS predictor index */
209     if( decControl->nChannelsInternal == 2 ) {
210         if(   lostFlag == FLAG_DECODE_NORMAL ||
211             ( lostFlag == FLAG_DECODE_LBRR && channel_state[ 0 ].LBRR_flags[ channel_state[ 0 ].nFramesDecoded ] == 1 ) )
212         {
213             silk_stereo_decode_pred( psRangeDec, MS_pred_Q13 );
214             /* For LBRR data, decode mid-only flag only if side-channel's LBRR flag is false */
215             if( ( lostFlag == FLAG_DECODE_NORMAL && channel_state[ 1 ].VAD_flags[ channel_state[ 0 ].nFramesDecoded ] == 0 ) ||
216                 ( lostFlag == FLAG_DECODE_LBRR && channel_state[ 1 ].LBRR_flags[ channel_state[ 0 ].nFramesDecoded ] == 0 ) )
217             {
218                 silk_stereo_decode_mid_only( psRangeDec, &decode_only_middle );
219             } else {
220                 decode_only_middle = 0;
221             }
222         } else {
223             for( n = 0; n < 2; n++ ) {
224                 MS_pred_Q13[ n ] = psDec->sStereo.pred_prev_Q13[ n ];
225             }
226         }
227     }
228
229     /* Reset side channel decoder prediction memory for first frame with side coding */
230     if( decControl->nChannelsInternal == 2 && decode_only_middle == 0 && psDec->prev_decode_only_middle == 1 ) {
231         silk_memset( psDec->channel_state[ 1 ].outBuf, 0, sizeof(psDec->channel_state[ 1 ].outBuf) );
232         silk_memset( psDec->channel_state[ 1 ].sLPC_Q14_buf, 0, sizeof(psDec->channel_state[ 1 ].sLPC_Q14_buf) );
233         psDec->channel_state[ 1 ].lagPrev        = 100;
234         psDec->channel_state[ 1 ].LastGainIndex  = 10;
235         psDec->channel_state[ 1 ].prevSignalType = TYPE_NO_VOICE_ACTIVITY;
236         psDec->channel_state[ 1 ].first_frame_after_reset = 1;
237     }
238
239     /* Call decoder for one frame */
240     for( n = 0; n < decControl->nChannelsInternal; n++ ) {
241         if( n == 0 || ( ( lostFlag != FLAG_PACKET_LOST ? decode_only_middle : psDec->prev_decode_only_middle ) == 0 ) ) {
242             opus_int FrameIndex;
243             opus_int condCoding;
244
245             FrameIndex = channel_state[ 0 ].nFramesDecoded - n;
246             /* Use independent coding if no previous frame available */
247             if( FrameIndex <= 0 ) {
248                 condCoding = CODE_INDEPENDENTLY;
249             } else if( lostFlag == FLAG_DECODE_LBRR ) {
250                 condCoding = channel_state[ n ].LBRR_flags[ FrameIndex - 1 ] ? CODE_CONDITIONALLY : CODE_INDEPENDENTLY;
251             } else if( n > 0 && psDec->prev_decode_only_middle ) {
252                 /* If we skipped a side frame in this packet, we don't
253                    need LTP scaling; the LTP state is well-defined. */
254                 condCoding = CODE_INDEPENDENTLY_NO_LTP_SCALING;
255             } else {
256                 condCoding = CODE_CONDITIONALLY;
257             }
258             ret += silk_decode_frame( &channel_state[ n ], psRangeDec, &samplesOut1_tmp[ n ][ 2 + delay ], &nSamplesOutDec, lostFlag, condCoding);
259         } else {
260             silk_memset( &samplesOut1_tmp[ n ][ 2 + delay ], 0, nSamplesOutDec * sizeof( opus_int16 ) );
261         }
262         channel_state[ n ].nFramesDecoded++;
263     }
264
265     if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 2 ) {
266         /* Convert Mid/Side to Left/Right */
267         silk_stereo_MS_to_LR( &psDec->sStereo, &samplesOut1_tmp[ 0 ][ delay ], &samplesOut1_tmp[ 1 ][ delay ], MS_pred_Q13, channel_state[ 0 ].fs_kHz, nSamplesOutDec );
268     } else {
269         /* Buffering */
270         silk_memcpy( &samplesOut1_tmp[ 0 ][ delay ], psDec->sStereo.sMid, 2 * sizeof( opus_int16 ) );
271         silk_memcpy( psDec->sStereo.sMid, &samplesOut1_tmp[ 0 ][ nSamplesOutDec + delay ], 2 * sizeof( opus_int16 ) );
272     }
273
274     /* Number of output samples */
275     *nSamplesOut = silk_DIV32( nSamplesOutDec * decControl->API_sampleRate, silk_SMULBB( channel_state[ 0 ].fs_kHz, 1000 ) );
276
277     /* Set up pointers to temp buffers */
278     if( decControl->nChannelsAPI == 2 ) {
279         resample_out_ptr = samplesOut2_tmp;
280     } else {
281         resample_out_ptr = samplesOut;
282     }
283
284     for( n = 0; n < silk_min( decControl->nChannelsAPI, decControl->nChannelsInternal ); n++ ) {
285
286         silk_memcpy(&samplesOut1_tmp[ n ][ 1 ], &channel_state[ n ].delayBuf[ MAX_DECODER_DELAY - delay ], delay * sizeof(opus_int16));
287         /* Resample decoded signal to API_sampleRate */
288         ret += silk_resampler( &channel_state[ n ].resampler_state, resample_out_ptr, &samplesOut1_tmp[ n ][ 1 ], nSamplesOutDec );
289         silk_memcpy(channel_state[ n ].delayBuf, &samplesOut1_tmp[ n ][ 1 + nSamplesOutDec + delay - MAX_DECODER_DELAY ], MAX_DECODER_DELAY * sizeof(opus_int16));
290
291         /* Interleave if stereo output and stereo stream */
292         if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 2 ) {
293             for( i = 0; i < *nSamplesOut; i++ ) {
294                 samplesOut[ n + 2 * i ] = resample_out_ptr[ i ];
295             }
296         }
297     }
298
299     /* Create two channel output from mono stream */
300     if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 1 ) {
301         for( i = 0; i < *nSamplesOut; i++ ) {
302             samplesOut[ 0 + 2 * i ] = samplesOut[ 1 + 2 * i ] = resample_out_ptr[ i ];
303         }
304     }
305
306     /* Export pitch lag, measured at 48 kHz sampling rate */
307     if( channel_state[ 0 ].prevSignalType == TYPE_VOICED ) {
308         int mult_tab[ 3 ] = { 6, 4, 3 };
309         decControl->prevPitchLag = channel_state[ 0 ].lagPrev * mult_tab[ ( channel_state[ 0 ].fs_kHz - 8 ) >> 2 ];
310     } else {
311         decControl->prevPitchLag = 0;
312     }
313
314     if ( lostFlag != FLAG_PACKET_LOST ) {
315        psDec->prev_decode_only_middle = decode_only_middle;
316     }
317     return ret;
318 }
319
320 /* Getting table of contents for a packet */
321 opus_int silk_get_TOC(
322     const opus_uint8                     *payload,           /* I    Payload data                                */
323     const opus_int                       nBytesIn,           /* I:   Number of input bytes                       */
324     const opus_int                       nFramesPerPayload,  /* I:   Number of SILK frames per payload           */
325     silk_TOC_struct                 *Silk_TOC           /* O:   Type of content                             */
326 )
327 {
328     opus_int i, flags, ret = SILK_NO_ERROR;
329
330     if( nBytesIn < 1 ) {
331         return -1;
332     }
333     if( nFramesPerPayload < 0 || nFramesPerPayload > 3 ) {
334         return -1;
335     }
336
337     silk_memset( Silk_TOC, 0, sizeof( Silk_TOC ) );
338
339     /* For stereo, extract the flags for the mid channel */
340     flags = silk_RSHIFT( payload[ 0 ], 7 - nFramesPerPayload ) & ( silk_LSHIFT( 1, nFramesPerPayload + 1 ) - 1 );
341
342     Silk_TOC->inbandFECFlag = flags & 1;
343     for( i = nFramesPerPayload - 1; i >= 0 ; i-- ) {
344         flags = silk_RSHIFT( flags, 1 );
345         Silk_TOC->VADFlags[ i ] = flags & 1;
346         Silk_TOC->VADFlag |= flags & 1;
347     }
348
349     return ret;
350 }