opus_int silk_setup_resamplers(
silk_encoder_state_Fxx *psEnc, /* I/O */
opus_int fs_kHz /* I */
)
{
opus_int ret = SILK_NO_ERROR;
opus_int32 nSamples_temp;
if( psEnc->sCmn.fs_kHz != fs_kHz || psEnc->sCmn.prev_API_fs_Hz != psEnc->sCmn.API_fs_Hz )
{
if( psEnc->sCmn.fs_kHz == 0 ) {
/* Initialize the resampler for enc_API.c preparing resampling from API_fs_Hz to fs_kHz */
ret += silk_resampler_init( &psEnc->sCmn.resampler_state, psEnc->sCmn.API_fs_Hz, fs_kHz * 1000, 1 );
} else {
/* Allocate worst case space for temporary upsampling, 8 to 48 kHz, so a factor 6 */
opus_int16 x_buf_API_fs_Hz[ ( 2 * MAX_FRAME_LENGTH_MS + LA_SHAPE_MS ) * MAX_API_FS_KHZ ];
silk_resampler_state_struct temp_resampler_state;
#ifdef FIXED_POINT
opus_int16 *x_bufFIX = psEnc->x_buf;
#else
opus_int16 x_bufFIX[ 2 * MAX_FRAME_LENGTH + LA_SHAPE_MAX ];
#endif
nSamples_temp = silk_LSHIFT( psEnc->sCmn.frame_length, 1 ) + LA_SHAPE_MS * psEnc->sCmn.fs_kHz;
#ifndef FIXED_POINT
silk_float2short_array( x_bufFIX, psEnc->x_buf, nSamples_temp );
#endif
/* Initialize resampler for temporary resampling of x_buf data to API_fs_Hz */
ret += silk_resampler_init( &temp_resampler_state, silk_SMULBB( psEnc->sCmn.fs_kHz, 1000 ), psEnc->sCmn.API_fs_Hz, 0 );
/* Temporary resampling of x_buf data to API_fs_Hz */
ret += silk_resampler( &temp_resampler_state, x_buf_API_fs_Hz, x_bufFIX, nSamples_temp );
/* Calculate number of samples that has been temporarily upsampled */
nSamples_temp = silk_DIV32_16( nSamples_temp * psEnc->sCmn.API_fs_Hz, silk_SMULBB( psEnc->sCmn.fs_kHz, 1000 ) );
/* Initialize the resampler for enc_API.c preparing resampling from API_fs_Hz to fs_kHz */
ret += silk_resampler_init( &psEnc->sCmn.resampler_state, psEnc->sCmn.API_fs_Hz, silk_SMULBB( fs_kHz, 1000 ), 1 );
/* Correct resampler state by resampling buffered data from API_fs_Hz to fs_kHz */
ret += silk_resampler( &psEnc->sCmn.resampler_state, x_bufFIX, x_buf_API_fs_Hz, nSamples_temp );
#ifndef FIXED_POINT
silk_short2float_array( psEnc->x_buf, x_bufFIX, ( 2 * MAX_FRAME_LENGTH_MS + LA_SHAPE_MS ) * fs_kHz );
#endif
}
}
psEnc->sCmn.prev_API_fs_Hz = psEnc->sCmn.API_fs_Hz;
return ret;
}
static opus_int silk_setup_resamplers(
silk_encoder_state_Fxx *psEnc, /* I/O */
opus_int fs_kHz /* I */
)
{
opus_int ret = SILK_NO_ERROR;
SAVE_STACK;
if( psEnc->sCmn.fs_kHz != fs_kHz || psEnc->sCmn.prev_API_fs_Hz != psEnc->sCmn.API_fs_Hz )
{
if( psEnc->sCmn.fs_kHz == 0 ) {
/* Initialize the resampler for enc_API.c preparing resampling from API_fs_Hz to fs_kHz */
ret += silk_resampler_init( &psEnc->sCmn.resampler_state, psEnc->sCmn.API_fs_Hz, fs_kHz * 1000, 1 );
} else {
VARDECL( opus_int16, x_buf_API_fs_Hz );
VARDECL( silk_resampler_state_struct, temp_resampler_state );
#ifdef OPUS_FIXED_POINT
opus_int16 *x_bufFIX = psEnc->x_buf;
#else
VARDECL( opus_int16, x_bufFIX );
opus_int32 new_buf_samples;
#endif
opus_int32 api_buf_samples;
opus_int32 old_buf_samples;
opus_int32 buf_length_ms;
buf_length_ms = silk_LSHIFT( psEnc->sCmn.nb_subfr * 5, 1 ) + LA_SHAPE_MS;
old_buf_samples = buf_length_ms * psEnc->sCmn.fs_kHz;
#ifndef OPUS_FIXED_POINT
new_buf_samples = buf_length_ms * fs_kHz;
ALLOC( x_bufFIX, silk_max( old_buf_samples, new_buf_samples ),
opus_int16 );
silk_float2short_array( x_bufFIX, psEnc->x_buf, old_buf_samples );
#endif
/* Initialize resampler for temporary resampling of x_buf data to API_fs_Hz */
ALLOC( temp_resampler_state, 1, silk_resampler_state_struct );
ret += silk_resampler_init( temp_resampler_state, silk_SMULBB( psEnc->sCmn.fs_kHz, 1000 ), psEnc->sCmn.API_fs_Hz, 0 );
/* Calculate number of samples to temporarily upsample */
api_buf_samples = buf_length_ms * silk_DIV32_16( psEnc->sCmn.API_fs_Hz, 1000 );
/* Temporary resampling of x_buf data to API_fs_Hz */
ALLOC( x_buf_API_fs_Hz, api_buf_samples, opus_int16 );
ret += silk_resampler( temp_resampler_state, x_buf_API_fs_Hz, x_bufFIX, old_buf_samples );
/* Initialize the resampler for enc_API.c preparing resampling from API_fs_Hz to fs_kHz */
ret += silk_resampler_init( &psEnc->sCmn.resampler_state, psEnc->sCmn.API_fs_Hz, silk_SMULBB( fs_kHz, 1000 ), 1 );
/* Correct resampler state by resampling buffered data from API_fs_Hz to fs_kHz */
ret += silk_resampler( &psEnc->sCmn.resampler_state, x_bufFIX, x_buf_API_fs_Hz, api_buf_samples );
#ifndef OPUS_FIXED_POINT
silk_short2float_array( psEnc->x_buf, x_bufFIX, new_buf_samples);
#endif
}
}
psEnc->sCmn.prev_API_fs_Hz = psEnc->sCmn.API_fs_Hz;
RESTORE_STACK;
return ret;
}
/* 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, nSamplesToBufferMax, nBlocksOf10ms;
opus_int nSamplesFromInput = 0, nSamplesFromInputMax;
opus_int speech_act_thr_for_switch_Q8;
opus_int32 TargetRate_bps, MStargetRates_bps[ 2 ], channelRate_bps, LBRR_symbol, sum;
silk_encoder *psEnc = ( silk_encoder * )encState;
VARDECL( opus_int16, buf );
opus_int transition, curr_block, tot_blocks;
SAVE_STACK;
if (encControl->reducedDependency)
{
psEnc->state_Fxx[0].sCmn.first_frame_after_reset = 1;
psEnc->state_Fxx[1].sCmn.first_frame_after_reset = 1;
}
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 ) {
silk_assert( 0 );
RESTORE_STACK;
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 ], psEnc->state_Fxx[ 0 ].sCmn.arch );
silk_memset( psEnc->sStereo.pred_prev_Q13, 0, sizeof( psEnc->sStereo.pred_prev_Q13 ) );
silk_memset( psEnc->sStereo.sSide, 0, sizeof( psEnc->sStereo.sSide ) );
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 ) {
silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, sizeof( silk_resampler_state_struct ) );
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 ) {
silk_assert( 0 );
RESTORE_STACK;
return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES;
}
/* Reset Encoder */
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
ret = silk_init_encoder( &psEnc->state_Fxx[ n ], psEnc->state_Fxx[ n ].sCmn.arch );
silk_assert( !ret );
}
tmp_payloadSize_ms = encControl->payloadSize_ms;
encControl->payloadSize_ms = 10;
tmp_complexity = encControl->complexity;
encControl->complexity = 0;
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0;
psEnc->state_Fxx[ n ].sCmn.prefillFlag = 1;
}
} else {
/* Only accept input lengths that are a multiple of 10 ms */
if( nBlocksOf10ms * encControl->API_sampleRate != 100 * nSamplesIn || nSamplesIn < 0 ) {
silk_assert( 0 );
RESTORE_STACK;
return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES;
}
/* Make sure no more than one packet can be produced */
if( 1000 * (opus_int32)nSamplesIn > encControl->payloadSize_ms * encControl->API_sampleRate ) {
silk_assert( 0 );
RESTORE_STACK;
return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES;
}
}
TargetRate_bps = silk_RSHIFT32( encControl->bitRate, encControl->nChannelsInternal - 1 );
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
/* 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 );
RESTORE_STACK;
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 );
/* Input buffering/resampling and encoding */
nSamplesToBufferMax =
10 * nBlocksOf10ms * psEnc->state_Fxx[ 0 ].sCmn.fs_kHz;
nSamplesFromInputMax =
silk_DIV32_16( nSamplesToBufferMax *
psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz,
psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 );
ALLOC( buf, nSamplesFromInputMax, opus_int16 );
while( 1 ) {
nSamplesToBuffer = psEnc->state_Fxx[ 0 ].sCmn.frame_length - psEnc->state_Fxx[ 0 ].sCmn.inputBufIx;
nSamplesToBuffer = silk_min( nSamplesToBuffer, nSamplesToBufferMax );
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;
nSamplesToBuffer = psEnc->state_Fxx[ 1 ].sCmn.frame_length - psEnc->state_Fxx[ 1 ].sCmn.inputBufIx;
nSamplesToBuffer = silk_min( nSamplesToBuffer, 10 * nBlocksOf10ms * psEnc->state_Fxx[ 1 ].sCmn.fs_kHz );
for( n = 0; n < nSamplesFromInput; n++ ) {
buf[ n ] = samplesIn[ 2 * n + 1 ];
}
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++ ) {
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 ], buf, nSamplesFromInput );
psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer;
}
samplesIn += nSamplesFromInput * encControl->nChannelsAPI;
nSamplesIn -= nSamplesFromInput;
/* Default */
psEnc->allowBandwidthSwitch = 0;
/* Silk encoder */
if( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx >= psEnc->state_Fxx[ 0 ].sCmn.frame_length ) {
/* Enough data in input buffer, so encode */
silk_assert( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx == psEnc->state_Fxx[ 0 ].sCmn.frame_length );
silk_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inputBufIx == psEnc->state_Fxx[ 1 ].sCmn.frame_length );
/* Deal with LBRR data */
if( psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == 0 && !prefillFlag ) {
/* Create space at start of payload for VAD and FEC flags */
opus_uint8 iCDF[ 2 ] = { 0, 0 };
iCDF[ 0 ] = 256 - silk_RSHIFT( 256, ( psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket + 1 ) * encControl->nChannelsInternal );
ec_enc_icdf( psRangeEnc, 0, iCDF, 8 );
/* Encode any LBRR data from previous packet */
/* Encode LBRR flags */
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
LBRR_symbol = 0;
for( i = 0; i < psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket; i++ ) {
LBRR_symbol |= silk_LSHIFT( psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ], i );
}
psEnc->state_Fxx[ n ].sCmn.LBRR_flag = LBRR_symbol > 0 ? 1 : 0;
if( LBRR_symbol && psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket > 1 ) {
ec_enc_icdf( psRangeEnc, LBRR_symbol - 1, silk_LBRR_flags_iCDF_ptr[ psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket - 2 ], 8 );
}
}
/* Code LBRR indices and excitation signals */
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 */
if( psEnc->state_Fxx[ 1 ].sCmn.LBRR_flags[ i ] == 0 ) {
silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ i ] );
}
}
/* 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 );
}
}
}
/* Reset LBRR flags */
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
silk_memset( psEnc->state_Fxx[ n ].sCmn.LBRR_flags, 0, sizeof( psEnc->state_Fxx[ n ].sCmn.LBRR_flags ) );
}
psEnc->nBitsUsedLBRR = ec_tell( psRangeEnc );
}
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 bits used for LBRR */
if( !prefillFlag ) {
nBits -= psEnc->nBitsUsedLBRR;
}
/* Divide by number of uncoded frames left in packet */
nBits = silk_DIV32_16( nBits, psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket );
/* Convert to bits/second */
if( encControl->payloadSize_ms == 10 ) {
TargetRate_bps = silk_SMULBB( nBits, 100 );
} else {
TargetRate_bps = silk_SMULBB( nBits, 50 );
}
/* Subtract fraction of bits in excess of target in previous frames and packets */
TargetRate_bps -= silk_DIV32_16( silk_MUL( psEnc->nBitsExceeded, 1000 ), BITRESERVOIR_DECAY_TIME_MS );
if( !prefillFlag && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded > 0 ) {
/* Compare actual vs target bits so far in this packet */
opus_int32 bitsBalance = ec_tell( psRangeEnc ) - psEnc->nBitsUsedLBRR - nBits * psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded;
TargetRate_bps -= silk_DIV32_16( silk_MUL( bitsBalance, 1000 ), BITRESERVOIR_DECAY_TIME_MS );
}
/* Never exceed input bitrate */
TargetRate_bps = silk_LIMIT( TargetRate_bps, encControl->bitRate, 5000 );
/* Convert Left/Right to Mid/Side */
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, encControl->toMono,
psEnc->state_Fxx[ 0 ].sCmn.fs_kHz, psEnc->state_Fxx[ 0 ].sCmn.frame_length );
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 ] );
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 );
/* 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 && 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++ ) {
flags = silk_LSHIFT( flags, 1 );
flags |= psEnc->state_Fxx[ n ].sCmn.VAD_flags[ i ];
}
flags = silk_LSHIFT( flags, 1 );
flags |= psEnc->state_Fxx[ n ].sCmn.LBRR_flag;
}
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 ) ) {
*nBytesOut = 0;
}
psEnc->nBitsExceeded += *nBytesOut * 8;
psEnc->nBitsExceeded -= silk_DIV32_16( silk_MUL( encControl->bitRate, encControl->payloadSize_ms ), 1000 );
psEnc->nBitsExceeded = silk_LIMIT( psEnc->nBitsExceeded, 0, 10000 );
/* Update flag indicating if bandwidth switching is allowed */
speech_act_thr_for_switch_Q8 = (opus_int) silk_SMLAWB( SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ),
SILK_FIX_CONST( ( 1 - SPEECH_ACTIVITY_DTX_THRES ) / MAX_BANDWIDTH_SWITCH_DELAY_MS, 16 + 8 ), psEnc->timeSinceSwitchAllowed_ms );
if( psEnc->state_Fxx[ 0 ].sCmn.speech_activity_Q8 < speech_act_thr_for_switch_Q8 ) {
psEnc->allowBandwidthSwitch = 1;
psEnc->timeSinceSwitchAllowed_ms = 0;
} else {
psEnc->allowBandwidthSwitch = 0;
psEnc->timeSinceSwitchAllowed_ms += encControl->payloadSize_ms;
}
}
if( nSamplesIn == 0 ) {
break;
}
} 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 = encControl->toMono ? 0 : psEnc->sStereo.smth_width_Q14;
if( prefillFlag ) {
encControl->payloadSize_ms = tmp_payloadSize_ms;
encControl->complexity = tmp_complexity;
for( n = 0; n < encControl->nChannelsInternal; n++ ) {
psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0;
psEnc->state_Fxx[ n ].sCmn.prefillFlag = 0;
}
}
RESTORE_STACK;
return ret;
}
