Fixes a few minor issues (no bit-stream change)
[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(                         /* O    Returns error code                              */
50     opus_int                        *decSizeBytes       /* O    Number of bytes in SILK decoder state           */
51 )
52 {
53     opus_int ret = SILK_NO_ERROR;
54
55     *decSizeBytes = sizeof( silk_decoder );
56
57     return ret;
58 }
59
60 /* Reset decoder state */
61 opus_int silk_InitDecoder(                              /* O    Returns error code                              */
62     void                            *decState           /* I/O  State                                           */
63 )
64 {
65     opus_int n, ret = SILK_NO_ERROR;
66     silk_decoder_state *channel_state = ((silk_decoder *)decState)->channel_state;
67
68     for( n = 0; n < DECODER_NUM_CHANNELS; n++ ) {
69         ret  = silk_init_decoder( &channel_state[ n ] );
70     }
71
72     return ret;
73 }
74
75 /* Decode a frame */
76 opus_int silk_Decode(                                   /* O    Returns error code                              */
77     void*                           decState,           /* I/O  State                                           */
78     silk_DecControlStruct*          decControl,         /* I/O  Control Structure                               */
79     opus_int                        lostFlag,           /* I    0: no loss, 1 loss, 2 decode fec                */
80     opus_int                        newPacketFlag,      /* I    Indicates first decoder call for this packet    */
81     ec_dec                          *psRangeDec,        /* I/O  Compressor data structure                       */
82     opus_int16                      *samplesOut,        /* O    Decoded output speech vector                    */
83     opus_int32                      *nSamplesOut        /* O    Number of samples decoded                       */
84 )
85 {
86     opus_int   i, n, decode_only_middle = 0, ret = SILK_NO_ERROR;
87     opus_int32 nSamplesOutDec, LBRR_symbol;
88     opus_int16 samplesOut1_tmp[ 2 ][ MAX_FS_KHZ * MAX_FRAME_LENGTH_MS + 2 ];
89     opus_int16 samplesOut2_tmp[ MAX_API_FS_KHZ * MAX_FRAME_LENGTH_MS ];
90     opus_int32 MS_pred_Q13[ 2 ] = { 0 };
91     opus_int16 *resample_out_ptr;
92     silk_decoder *psDec = ( silk_decoder * )decState;
93     silk_decoder_state *channel_state = psDec->channel_state;
94     opus_int has_side;
95
96     /**********************************/
97     /* Test if first frame in payload */
98     /**********************************/
99     if( newPacketFlag ) {
100         for( n = 0; n < decControl->nChannelsInternal; n++ ) {
101             channel_state[ n ].nFramesDecoded = 0;  /* Used to count frames in packet */
102         }
103     }
104
105     /* If Mono -> Stereo transition in bitstream: init state of second channel */
106     if( decControl->nChannelsInternal > psDec->nChannelsInternal ) {
107         ret += silk_init_decoder( &channel_state[ 1 ] );
108     }
109
110     if( channel_state[ 0 ].nFramesDecoded == 0 ) {
111         for( n = 0; n < decControl->nChannelsInternal; n++ ) {
112             opus_int fs_kHz_dec;
113             if( decControl->payloadSize_ms == 0 ) {
114                 /* Assuming packet loss, use 10 ms */
115                 channel_state[ n ].nFramesPerPacket = 1;
116                 channel_state[ n ].nb_subfr = 2;
117             } else if( decControl->payloadSize_ms == 10 ) {
118                 channel_state[ n ].nFramesPerPacket = 1;
119                 channel_state[ n ].nb_subfr = 2;
120             } else if( decControl->payloadSize_ms == 20 ) {
121                 channel_state[ n ].nFramesPerPacket = 1;
122                 channel_state[ n ].nb_subfr = 4;
123             } else if( decControl->payloadSize_ms == 40 ) {
124                 channel_state[ n ].nFramesPerPacket = 2;
125                 channel_state[ n ].nb_subfr = 4;
126             } else if( decControl->payloadSize_ms == 60 ) {
127                 channel_state[ n ].nFramesPerPacket = 3;
128                 channel_state[ n ].nb_subfr = 4;
129             } else {
130                 silk_assert( 0 );
131                 return SILK_DEC_INVALID_FRAME_SIZE;
132             }
133             fs_kHz_dec = ( decControl->internalSampleRate >> 10 ) + 1;
134             if( fs_kHz_dec != 8 && fs_kHz_dec != 12 && fs_kHz_dec != 16 ) {
135                 silk_assert( 0 );
136                 return SILK_DEC_INVALID_SAMPLING_FREQUENCY;
137             }
138             ret += silk_decoder_set_fs( &channel_state[ n ], fs_kHz_dec, decControl->API_sampleRate );
139         }
140     }
141
142     if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 2 && ( psDec->nChannelsAPI == 1 || psDec->nChannelsInternal == 1 ) ) {
143         silk_memset( psDec->sStereo.pred_prev_Q13, 0, sizeof( psDec->sStereo.pred_prev_Q13 ) );
144         silk_memset( psDec->sStereo.sSide, 0, sizeof( psDec->sStereo.sSide ) );
145         silk_memcpy( &channel_state[ 1 ].resampler_state, &channel_state[ 0 ].resampler_state, sizeof( silk_resampler_state_struct ) );
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     if( lostFlag == FLAG_DECODE_NORMAL ) {
240         has_side = !decode_only_middle;
241     } else {
242         has_side = !psDec->prev_decode_only_middle
243               || (decControl->nChannelsInternal == 2 && lostFlag == FLAG_DECODE_LBRR && channel_state[1].LBRR_flags[ channel_state[1].nFramesDecoded ] == 1 );
244     }
245     /* Call decoder for one frame */
246     for( n = 0; n < decControl->nChannelsInternal; n++ ) {
247         if( n == 0 || has_side ) {
248             opus_int FrameIndex;
249             opus_int condCoding;
250
251             FrameIndex = channel_state[ 0 ].nFramesDecoded - n;
252             /* Use independent coding if no previous frame available */
253             if( FrameIndex <= 0 ) {
254                 condCoding = CODE_INDEPENDENTLY;
255             } else if( lostFlag == FLAG_DECODE_LBRR ) {
256                 condCoding = channel_state[ n ].LBRR_flags[ FrameIndex - 1 ] ? CODE_CONDITIONALLY : CODE_INDEPENDENTLY;
257             } else if( n > 0 && psDec->prev_decode_only_middle ) {
258                 /* If we skipped a side frame in this packet, we don't
259                    need LTP scaling; the LTP state is well-defined. */
260                 condCoding = CODE_INDEPENDENTLY_NO_LTP_SCALING;
261             } else {
262                 condCoding = CODE_CONDITIONALLY;
263             }
264             ret += silk_decode_frame( &channel_state[ n ], psRangeDec, &samplesOut1_tmp[ n ][ 2 ], &nSamplesOutDec, lostFlag, condCoding);
265         } else {
266             silk_memset( &samplesOut1_tmp[ n ][ 2 ], 0, nSamplesOutDec * sizeof( opus_int16 ) );
267         }
268         channel_state[ n ].nFramesDecoded++;
269     }
270
271     if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 2 ) {
272         /* Convert Mid/Side to Left/Right */
273         silk_stereo_MS_to_LR( &psDec->sStereo, samplesOut1_tmp[ 0 ], samplesOut1_tmp[ 1 ], MS_pred_Q13, channel_state[ 0 ].fs_kHz, nSamplesOutDec );
274     } else {
275         /* Buffering */
276         silk_memcpy( samplesOut1_tmp[ 0 ], psDec->sStereo.sMid, 2 * sizeof( opus_int16 ) );
277         silk_memcpy( psDec->sStereo.sMid, &samplesOut1_tmp[ 0 ][ nSamplesOutDec ], 2 * sizeof( opus_int16 ) );
278     }
279
280     /* Number of output samples */
281     *nSamplesOut = silk_DIV32( nSamplesOutDec * decControl->API_sampleRate, silk_SMULBB( channel_state[ 0 ].fs_kHz, 1000 ) );
282
283     /* Set up pointers to temp buffers */
284     if( decControl->nChannelsAPI == 2 ) {
285         resample_out_ptr = samplesOut2_tmp;
286     } else {
287         resample_out_ptr = samplesOut;
288     }
289
290     for( n = 0; n < silk_min( decControl->nChannelsAPI, decControl->nChannelsInternal ); n++ ) {
291
292         /* Resample decoded signal to API_sampleRate */
293         ret += silk_resampler( &channel_state[ n ].resampler_state, resample_out_ptr, &samplesOut1_tmp[ n ][ 1 ], nSamplesOutDec );
294
295         /* Interleave if stereo output and stereo stream */
296         if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 2 ) {
297             for( i = 0; i < *nSamplesOut; i++ ) {
298                 samplesOut[ n + 2 * i ] = resample_out_ptr[ i ];
299             }
300         }
301     }
302
303     /* Create two channel output from mono stream */
304     if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 1 ) {
305         for( i = 0; i < *nSamplesOut; i++ ) {
306             samplesOut[ 0 + 2 * i ] = samplesOut[ 1 + 2 * i ] = resample_out_ptr[ i ];
307         }
308     }
309
310     /* Export pitch lag, measured at 48 kHz sampling rate */
311     if( channel_state[ 0 ].prevSignalType == TYPE_VOICED ) {
312         int mult_tab[ 3 ] = { 6, 4, 3 };
313         decControl->prevPitchLag = channel_state[ 0 ].lagPrev * mult_tab[ ( channel_state[ 0 ].fs_kHz - 8 ) >> 2 ];
314     } else {
315         decControl->prevPitchLag = 0;
316     }
317
318     if( lostFlag == FLAG_PACKET_LOST ) {
319        /* On packet loss, remove the gain clamping to prevent having the energy "bounce back"
320           if we lose packets when the energy is going down */
321        for ( i = 0; i < psDec->nChannelsInternal; i++ )
322           psDec->channel_state[ i ].LastGainIndex = 10;
323     } else {
324        psDec->prev_decode_only_middle = decode_only_middle;
325     }
326     return ret;
327 }
328
329 /* Getting table of contents for a packet */
330 opus_int silk_get_TOC(
331     const opus_uint8                *payload,           /* I    Payload data                                */
332     const opus_int                  nBytesIn,           /* I    Number of input bytes                       */
333     const opus_int                  nFramesPerPayload,  /* I    Number of SILK frames per payload           */
334     silk_TOC_struct                 *Silk_TOC           /* O    Type of content                             */
335 )
336 {
337     opus_int i, flags, ret = SILK_NO_ERROR;
338
339     if( nBytesIn < 1 ) {
340         return -1;
341     }
342     if( nFramesPerPayload < 0 || nFramesPerPayload > 3 ) {
343         return -1;
344     }
345
346     silk_memset( Silk_TOC, 0, sizeof( Silk_TOC ) );
347
348     /* For stereo, extract the flags for the mid channel */
349     flags = silk_RSHIFT( payload[ 0 ], 7 - nFramesPerPayload ) & ( silk_LSHIFT( 1, nFramesPerPayload + 1 ) - 1 );
350
351     Silk_TOC->inbandFECFlag = flags & 1;
352     for( i = nFramesPerPayload - 1; i >= 0 ; i-- ) {
353         flags = silk_RSHIFT( flags, 1 );
354         Silk_TOC->VADFlags[ i ] = flags & 1;
355         Silk_TOC->VADFlag |= flags & 1;
356     }
357
358     return ret;
359 }