Use configure.ac provided soname versioning.
[opus.git] / silk / enc_API.c
index 29d7568..c0143fd 100644 (file)
@@ -1,28 +1,28 @@
 /***********************************************************************
 Copyright (c) 2006-2011, Skype Limited. All rights reserved.
 Redistribution and use in source and binary forms, with or without
-modification, (subject to the limitations in the disclaimer below)
-are permitted provided that the following conditions are met:
+modification, are permitted provided that the following conditions
+are met:
 - Redistributions of source code must retain the above copyright notice,
 this list of conditions and the following disclaimer.
 - Redistributions in binary form must reproduce the above copyright
 notice, this list of conditions and the following disclaimer in the
 documentation and/or other materials provided with the distribution.
-- Neither the name of Skype Limited, nor the names of specific
-contributors, may be used to endorse or promote products derived from
-this software without specific prior written permission.
-NO EXPRESS OR IMPLIED LICENSES TO ANY PARTY'S PATENT RIGHTS ARE GRANTED
-BY THIS LICENSE. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
-CONTRIBUTORS ''AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING,
-BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
-FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
-COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
-INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
-NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
-USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
-ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+- Neither the name of Internet Society, IETF or IETF Trust, nor the 
+names of specific contributors, may be used to endorse or promote
+products derived from this software without specific prior written
+permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS”
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+POSSIBILITY OF SUCH DAMAGE.
 ***********************************************************************/
 
 #ifdef HAVE_CONFIG_H
@@ -44,7 +44,9 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 /* Encoder functions                    */
 /****************************************/
 
-opus_int silk_Get_Encoder_Size( int *encSizeBytes )
+opus_int silk_Get_Encoder_Size(                         /* O    Returns error code                              */
+    opus_int                        *encSizeBytes       /* O    Number of bytes in SILK encoder state           */
+)
 {
     opus_int ret = SILK_NO_ERROR;
 
@@ -56,9 +58,9 @@ opus_int silk_Get_Encoder_Size( int *encSizeBytes )
 /*************************/
 /* Init or Reset encoder */
 /*************************/
-opus_int silk_InitEncoder(
-    void                            *encState,          /* I/O: State                                           */
-    silk_EncControlStruct           *encStatus          /* O:   Control structure                               */
+opus_int silk_InitEncoder(                              /* O    Returns error code                              */
+    void                            *encState,          /* I/O  State                                           */
+    silk_EncControlStruct           *encStatus          /* O    Encoder Status                                  */
 )
 {
     silk_encoder *psEnc;
@@ -88,9 +90,9 @@ opus_int silk_InitEncoder(
 /***************************************/
 /* Read control structure from encoder */
 /***************************************/
-opus_int silk_QueryEncoder(
-    const void *encState,                       /* I:   State Vector                                    */
-    silk_EncControlStruct *encStatus            /* O:   Control Structure                               */
+opus_int silk_QueryEncoder(                             /* O    Returns error code                              */
+    const void                      *encState,          /* I    State                                           */
+    silk_EncControlStruct           *encStatus          /* O    Encoder Status                                  */
 )
 {
     opus_int ret = SILK_NO_ERROR;
@@ -119,25 +121,31 @@ opus_int silk_QueryEncoder(
     return ret;
 }
 
+
 /**************************/
 /* Encode frame with Silk */
 /**************************/
-opus_int silk_Encode(
-    void                                *encState,      /* I/O: State                                           */
-    silk_EncControlStruct               *encControl,    /* I:   Control structure                               */
-    const opus_int16                     *samplesIn,     /* I:   Speech sample input vector                      */
-    opus_int                             nSamplesIn,     /* I:   Number of samples in input vector               */
-    ec_enc                              *psRangeEnc,    /* I/O  Compressor data structure                       */
-    opus_int                             *nBytesOut,     /* I/O: Number of bytes in payload (input: Max bytes)   */
-    const opus_int                       prefillFlag     /* I:   Flag to indicate prefilling buffers; no coding  */
+/* Note: if prefillFlag is set, the input must contain 10 ms of audio, irrespective of what                     */
+/* encControl->payloadSize_ms is set to                                                                         */
+opus_int silk_Encode(                                   /* O    Returns error code                              */
+    void                            *encState,          /* I/O  State                                           */
+    silk_EncControlStruct           *encControl,        /* I    Control status                                  */
+    const opus_int16                *samplesIn,         /* I    Speech sample input vector                      */
+    opus_int                        nSamplesIn,         /* I    Number of samples in input vector               */
+    ec_enc                          *psRangeEnc,        /* I/O  Compressor data structure                       */
+    opus_int32                      *nBytesOut,         /* I/O  Number of bytes in payload (input: Max bytes)   */
+    const opus_int                  prefillFlag         /* I    Flag to indicate prefilling buffers no coding   */
 )
 {
     opus_int   n, i, nBits, flags, tmp_payloadSize_ms = 0, tmp_complexity = 0, ret = 0;
     opus_int   nSamplesToBuffer, nBlocksOf10ms, nSamplesFromInput = 0;
     opus_int   speech_act_thr_for_switch_Q8;
-    opus_int32 TargetRate_bps, MStargetRates_bps[ 2 ], channelRate_bps, LBRR_symbol;
+    opus_int32 TargetRate_bps, MStargetRates_bps[ 2 ], channelRate_bps, LBRR_symbol, sum;
     silk_encoder *psEnc = ( silk_encoder * )encState;
     opus_int16 buf[ MAX_FRAME_LENGTH_MS * MAX_API_FS_KHZ ];
+    opus_int transition, curr_block, tot_blocks;
+
+    psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded = psEnc->state_Fxx[ 1 ].sCmn.nFramesEncoded = 0;
 
     /* Check values in encoder control structure */
     if( ( ret = check_control_input( encControl ) != 0 ) ) {
@@ -145,12 +153,17 @@ opus_int silk_Encode(
         return ret;
     }
 
+    encControl->switchReady = 0;
+
     if( encControl->nChannelsInternal > psEnc->nChannelsInternal ) {
         /* Mono -> Stereo transition: init state of second channel and stereo state */
         ret += silk_init_encoder( &psEnc->state_Fxx[ 1 ] );
         silk_memset( psEnc->sStereo.pred_prev_Q13, 0, sizeof( psEnc->sStereo.pred_prev_Q13 ) );
         silk_memset( psEnc->sStereo.sSide, 0, sizeof( psEnc->sStereo.sSide ) );
-        silk_memset( psEnc->sStereo.mid_side_amp_Q0, 0, sizeof( psEnc->sStereo.mid_side_amp_Q0 ) );
+        psEnc->sStereo.mid_side_amp_Q0[ 0 ] = 0;
+        psEnc->sStereo.mid_side_amp_Q0[ 1 ] = 1;
+        psEnc->sStereo.mid_side_amp_Q0[ 2 ] = 0;
+        psEnc->sStereo.mid_side_amp_Q0[ 3 ] = 1;
         psEnc->sStereo.width_prev_Q14 = 0;
         psEnc->sStereo.smth_width_Q14 = SILK_FIX_CONST( 1, 14 );
         if( psEnc->nChannelsAPI == 2 ) {
@@ -158,10 +171,15 @@ opus_int silk_Encode(
             silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.In_HP_State,     &psEnc->state_Fxx[ 0 ].sCmn.In_HP_State,     sizeof( psEnc->state_Fxx[ 1 ].sCmn.In_HP_State ) );
         }
     }
+
+    transition = (encControl->payloadSize_ms != psEnc->state_Fxx[ 0 ].sCmn.PacketSize_ms) || (psEnc->nChannelsInternal != encControl->nChannelsInternal);
+
     psEnc->nChannelsAPI = encControl->nChannelsAPI;
     psEnc->nChannelsInternal = encControl->nChannelsInternal;
 
     nBlocksOf10ms = silk_DIV32( 100 * nSamplesIn, encControl->API_sampleRate );
+    tot_blocks = ( nBlocksOf10ms > 1 ) ? nBlocksOf10ms >> 1 : 1;
+    curr_block = 0;
     if( prefillFlag ) {
         /* Only accept input length of 10 ms */
         if( nBlocksOf10ms != 1 ) {
@@ -200,12 +218,18 @@ opus_int silk_Encode(
 
     TargetRate_bps = silk_RSHIFT32( encControl->bitRate, encControl->nChannelsInternal - 1 );
     for( n = 0; n < encControl->nChannelsInternal; n++ ) {
-        /* JMV: Force the side channel to the same rate as the mid. Is this the right way? */
-        int force_fs_kHz = (n==1) ? psEnc->state_Fxx[0].sCmn.fs_kHz : 0;
+        /* Force the side channel to the same rate as the mid */
+        opus_int force_fs_kHz = (n==1) ? psEnc->state_Fxx[0].sCmn.fs_kHz : 0;
         if( ( ret = silk_control_encoder( &psEnc->state_Fxx[ n ], encControl, TargetRate_bps, psEnc->allowBandwidthSwitch, n, force_fs_kHz ) ) != 0 ) {
             silk_assert( 0 );
             return ret;
         }
+        if( psEnc->state_Fxx[n].sCmn.first_frame_after_reset || transition ) {
+            for( i = 0; i < psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket; i++ ) {
+                psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ] = 0;
+            }
+        }
+        psEnc->state_Fxx[ n ].sCmn.inDTX = psEnc->state_Fxx[ n ].sCmn.useDTX;
     }
     silk_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == psEnc->state_Fxx[ 1 ].sCmn.fs_kHz );
 
@@ -216,9 +240,15 @@ opus_int silk_Encode(
         nSamplesFromInput = silk_DIV32_16( nSamplesToBuffer * psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz, psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 );
         /* Resample and write to buffer */
         if( encControl->nChannelsAPI == 2 && encControl->nChannelsInternal == 2 ) {
+            opus_int id = psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded;
             for( n = 0; n < nSamplesFromInput; n++ ) {
                 buf[ n ] = samplesIn[ 2 * n ];
             }
+            /* Making sure to start both resamplers from the same state when switching from mono to stereo */
+            if( psEnc->nPrevChannelsInternal == 1 && id==0 ) {
+               silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, sizeof(psEnc->state_Fxx[ 1 ].sCmn.resampler_state));
+            }
+
             ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state,
                 &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
             psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer;
@@ -230,21 +260,35 @@ opus_int silk_Encode(
             }
             ret += silk_resampler( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state,
                 &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
+
             psEnc->state_Fxx[ 1 ].sCmn.inputBufIx += nSamplesToBuffer;
         } else if( encControl->nChannelsAPI == 2 && encControl->nChannelsInternal == 1 ) {
             /* Combine left and right channels before resampling */
             for( n = 0; n < nSamplesFromInput; n++ ) {
-                buf[ n ] = (opus_int16)silk_RSHIFT_ROUND( samplesIn[ 2 * n ] + samplesIn[ 2 * n + 1 ],  1 );
+                sum = samplesIn[ 2 * n ] + samplesIn[ 2 * n + 1 ];
+                buf[ n ] = (opus_int16)silk_RSHIFT_ROUND( sum,  1 );
             }
             ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state,
                 &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
+            /* On the first mono frame, average the results for the two resampler states  */
+            if( psEnc->nPrevChannelsInternal == 2 && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == 0 ) {
+               ret += silk_resampler( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state,
+                   &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
+               for( n = 0; n < psEnc->state_Fxx[ 0 ].sCmn.frame_length; n++ ) {
+                  psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx+n+2 ] =
+                        silk_RSHIFT(psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx+n+2 ]
+                                  + psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx+n+2 ], 1);
+               }
+            }
             psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer;
         } else {
             silk_assert( encControl->nChannelsAPI == 1 && encControl->nChannelsInternal == 1 );
+            silk_memcpy(buf, samplesIn, nSamplesFromInput*sizeof(opus_int16));
             ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state,
-                &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], samplesIn, nSamplesFromInput );
+                &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput );
             psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer;
         }
+
         samplesIn  += nSamplesFromInput * encControl->nChannelsAPI;
         nSamplesIn -= nSamplesFromInput;
 
@@ -281,6 +325,8 @@ opus_int silk_Encode(
                 for( i = 0; i < psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket; i++ ) {
                     for( n = 0; n < encControl->nChannelsInternal; n++ ) {
                         if( psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ] ) {
+                            opus_int condCoding;
+
                             if( encControl->nChannelsInternal == 2 && n == 0 ) {
                                 silk_stereo_encode_pred( psRangeEnc, psEnc->sStereo.predIx[ i ] );
                                 /* For LBRR data there's no need to code the mid-only flag if the side-channel LBRR flag is set */
@@ -288,7 +334,13 @@ opus_int silk_Encode(
                                     silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ i ] );
                                 }
                             }
-                            silk_encode_indices( &psEnc->state_Fxx[ n ].sCmn, psRangeEnc, i, 1 );
+                            /* Use conditional coding if previous frame available */
+                            if( i > 0 && psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i - 1 ] ) {
+                                condCoding = CODE_CONDITIONALLY;
+                            } else {
+                                condCoding = CODE_INDEPENDENTLY;
+                            }
+                            silk_encode_indices( &psEnc->state_Fxx[ n ].sCmn, psRangeEnc, i, 1, condCoding );
                             silk_encode_pulses( psRangeEnc, psEnc->state_Fxx[ n ].sCmn.indices_LBRR[i].signalType, psEnc->state_Fxx[ n ].sCmn.indices_LBRR[i].quantOffsetType,
                                 psEnc->state_Fxx[ n ].sCmn.pulses_LBRR[ i ], psEnc->state_Fxx[ n ].sCmn.frame_length );
                         }
@@ -301,13 +353,14 @@ opus_int silk_Encode(
                 }
             }
 
-            silk_HP_variable_cutoff( psEnc->state_Fxx, psEnc->nChannelsInternal );
+            silk_HP_variable_cutoff( psEnc->state_Fxx );
 
             /* Total target bits for packet */
             nBits = silk_DIV32_16( silk_MUL( encControl->bitRate, encControl->payloadSize_ms ), 1000 );
             /* Subtract half of the bits already used */
-            if (!prefillFlag)
+            if( !prefillFlag ) {
                 nBits -= ec_tell( psRangeEnc ) >> 1;
+            }
             /* Divide by number of uncoded frames left in packet */
             nBits = silk_DIV32_16( nBits, psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket - psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded );
             /* Convert to bits/second */
@@ -325,40 +378,95 @@ opus_int silk_Encode(
             if( encControl->nChannelsInternal == 2 ) {
                 silk_stereo_LR_to_MS( &psEnc->sStereo, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ 2 ], &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ 2 ],
                     psEnc->sStereo.predIx[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ], &psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ],
-                    MStargetRates_bps, TargetRate_bps, psEnc->state_Fxx[ 0 ].sCmn.speech_activity_Q8,
+                    MStargetRates_bps, TargetRate_bps, psEnc->state_Fxx[ 0 ].sCmn.speech_activity_Q8, encControl->toMono,
                     psEnc->state_Fxx[ 0 ].sCmn.fs_kHz, psEnc->state_Fxx[ 0 ].sCmn.frame_length );
-                if (!prefillFlag) {
+                if( psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] == 0 ) {
+                    /* Reset side channel encoder memory for first frame with side coding */
+                    if( psEnc->prev_decode_only_middle == 1 ) {
+                        silk_memset( &psEnc->state_Fxx[ 1 ].sShape,               0, sizeof( psEnc->state_Fxx[ 1 ].sShape ) );
+                        silk_memset( &psEnc->state_Fxx[ 1 ].sPrefilt,             0, sizeof( psEnc->state_Fxx[ 1 ].sPrefilt ) );
+                        silk_memset( &psEnc->state_Fxx[ 1 ].sCmn.sNSQ,            0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.sNSQ ) );
+                        silk_memset( psEnc->state_Fxx[ 1 ].sCmn.prev_NLSFq_Q15,   0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.prev_NLSFq_Q15 ) );
+                        silk_memset( &psEnc->state_Fxx[ 1 ].sCmn.sLP.In_LP_State, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.sLP.In_LP_State ) );
+                        psEnc->state_Fxx[ 1 ].sCmn.prevLag                 = 100;
+                        psEnc->state_Fxx[ 1 ].sCmn.sNSQ.lagPrev            = 100;
+                        psEnc->state_Fxx[ 1 ].sShape.LastGainIndex         = 10;
+                        psEnc->state_Fxx[ 1 ].sCmn.prevSignalType          = TYPE_NO_VOICE_ACTIVITY;
+                        psEnc->state_Fxx[ 1 ].sCmn.sNSQ.prev_gain_Q16      = 65536;
+                        psEnc->state_Fxx[ 1 ].sCmn.first_frame_after_reset = 1;
+                    }
+                    silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 1 ] );
+                } else {
+                    psEnc->state_Fxx[ 1 ].sCmn.VAD_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] = 0;
+                }
+                if( !prefillFlag ) {
                     silk_stereo_encode_pred( psRangeEnc, psEnc->sStereo.predIx[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] );
-                    silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] );
+                    if( psEnc->state_Fxx[ 1 ].sCmn.VAD_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] == 0 ) {
+                        silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] );
+                    }
                 }
             } else {
                 /* Buffering */
                 silk_memcpy( psEnc->state_Fxx[ 0 ].sCmn.inputBuf, psEnc->sStereo.sMid, 2 * sizeof( opus_int16 ) );
                 silk_memcpy( psEnc->sStereo.sMid, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.frame_length ], 2 * sizeof( opus_int16 ) );
             }
+            silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 0 ] );
 
             /* Encode */
             for( n = 0; n < encControl->nChannelsInternal; n++ ) {
+                opus_int maxBits, useCBR;
+
+                /* Handling rate constraints */
+                maxBits = encControl->maxBits;
+                if( tot_blocks == 2 && curr_block == 0 ) {
+                    maxBits = maxBits * 3 / 5;
+                } else if( tot_blocks == 3 ) {
+                    if( curr_block == 0 ) {
+                        maxBits = maxBits * 2 / 5;
+                    } else if( curr_block == 1 ) {
+                        maxBits = maxBits * 3 / 4;
+                    }
+                }
+                useCBR = encControl->useCBR && curr_block == tot_blocks - 1;
+
                 if( encControl->nChannelsInternal == 1 ) {
                     channelRate_bps = TargetRate_bps;
                 } else {
                     channelRate_bps = MStargetRates_bps[ n ];
+                    if( n == 0 && MStargetRates_bps[ 1 ] > 0 ) {
+                        useCBR = 0;
+                        /* Give mid up to 1/2 of the max bits for that frame */
+                        maxBits -= encControl->maxBits / ( tot_blocks * 2 );
+                    }
                 }
 
                 if( channelRate_bps > 0 ) {
+                    opus_int condCoding;
+
                     silk_control_SNR( &psEnc->state_Fxx[ n ].sCmn, channelRate_bps );
 
-                    if( ( ret = silk_encode_frame_Fxx( &psEnc->state_Fxx[ n ], nBytesOut, psRangeEnc ) ) != 0 ) {
+                    /* Use independent coding if no previous frame available */
+                    if( psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded - n <= 0 ) {
+                        condCoding = CODE_INDEPENDENTLY;
+                    } else if( n > 0 && psEnc->prev_decode_only_middle ) {
+                        /* If we skipped a side frame in this packet, we don't
+                           need LTP scaling; the LTP state is well-defined. */
+                        condCoding = CODE_INDEPENDENTLY_NO_LTP_SCALING;
+                    } else {
+                        condCoding = CODE_CONDITIONALLY;
+                    }
+                    if( ( ret = silk_encode_frame_Fxx( &psEnc->state_Fxx[ n ], nBytesOut, psRangeEnc, condCoding, maxBits, useCBR ) ) != 0 ) {
                         silk_assert( 0 );
                     }
                 }
-
                 psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0;
                 psEnc->state_Fxx[ n ].sCmn.inputBufIx = 0;
+                psEnc->state_Fxx[ n ].sCmn.nFramesEncoded++;
             }
+            psEnc->prev_decode_only_middle = psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded - 1 ];
 
             /* Insert VAD and FEC flags at beginning of bitstream */
-            if( *nBytesOut > 0 ) {
+            if( *nBytesOut > 0 && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket) {
                 flags = 0;
                 for( n = 0; n < encControl->nChannelsInternal; n++ ) {
                     for( i = 0; i < psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket; i++ ) {
@@ -368,8 +476,9 @@ opus_int silk_Encode(
                     flags  = silk_LSHIFT( flags, 1 );
                     flags |= psEnc->state_Fxx[ n ].sCmn.LBRR_flag;
                 }
-                if (!prefillFlag)
+                if( !prefillFlag ) {
                     ec_enc_patch_initial_bits( psRangeEnc, flags, ( psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket + 1 ) * encControl->nChannelsInternal );
+                }
 
                 /* Return zero bytes if all channels DTXed */
                 if( psEnc->state_Fxx[ 0 ].sCmn.inDTX && ( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inDTX ) ) {
@@ -398,12 +507,15 @@ opus_int silk_Encode(
         } else {
             break;
         }
+        curr_block++;
     }
 
+    psEnc->nPrevChannelsInternal = encControl->nChannelsInternal;
+
     encControl->allowBandwidthSwitch = psEnc->allowBandwidthSwitch;
     encControl->inWBmodeWithoutVariableLP = psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == 16 && psEnc->state_Fxx[ 0 ].sCmn.sLP.mode == 0;
     encControl->internalSampleRate = silk_SMULBB( psEnc->state_Fxx[ 0 ].sCmn.fs_kHz, 1000 );
-    encControl->stereoWidth_Q14 = psEnc->sStereo.width_prev_Q14;
+    encControl->stereoWidth_Q14 = encControl->toMono ? 0 : psEnc->sStereo.smth_width_Q14;
     if( prefillFlag ) {
         encControl->payloadSize_ms = tmp_payloadSize_ms;
         encControl->complexity = tmp_complexity;