void SKP_Silk_quant_LTP_gains_FLP(
SKP_float B[ NB_SUBFR * LTP_ORDER ], /* I/O (Un-)quantized LTP gains */
SKP_int cbk_index[ NB_SUBFR ], /* O Codebook index */
SKP_int *periodicity_index, /* O Periodicity index */
const SKP_float W[ NB_SUBFR*LTP_ORDER*LTP_ORDER ], /* I Error weights */
const SKP_float mu, /* I Mu value (R/D tradeoff) */
const SKP_int lowComplexity /* I Flag for low complexity */
)
{
SKP_int j, k, temp_idx[ NB_SUBFR ], cbk_size;
const SKP_uint16 *cdf_ptr;
const SKP_int16 *cl_ptr;
const SKP_int16 *cbk_ptr_Q14;
const SKP_float *b_ptr, *W_ptr;
SKP_float rate_dist_subfr, rate_dist, min_rate_dist;
/***************************************************/
/* Iterate over different codebooks with different */
/* rates/distortions, and choose best */
/***************************************************/
min_rate_dist = SKP_float_MAX;
for( k = 0; k < 3; k++ ) {
cdf_ptr = SKP_Silk_LTP_gain_CDF_ptrs[ k ];
cl_ptr = SKP_Silk_LTP_gain_BITS_Q6_ptrs[ k ];
cbk_ptr_Q14 = SKP_Silk_LTP_vq_ptrs_Q14[ k ];
cbk_size = SKP_Silk_LTP_vq_sizes[ k ];
/* Setup pointer to first subframe */
W_ptr = W;
b_ptr = B;
rate_dist = 0.0f;
for( j = 0; j < NB_SUBFR; j++ ) {
SKP_Silk_VQ_WMat_EC_FLP(
&temp_idx[ j ], /* O index of best codebook vector */
&rate_dist_subfr, /* O best weighted quantization error + mu * rate */
b_ptr, /* I input vector to be quantized */
W_ptr, /* I weighting matrix */
cbk_ptr_Q14, /* I codebook */
cl_ptr, /* I code length for each codebook vector */
mu, /* I tradeoff between weighted error and rate */
cbk_size /* I number of vectors in codebook */
);
rate_dist += rate_dist_subfr;
b_ptr += LTP_ORDER;
W_ptr += LTP_ORDER * LTP_ORDER;
}
if( rate_dist < min_rate_dist ) {
min_rate_dist = rate_dist;
SKP_memcpy( cbk_index, temp_idx, NB_SUBFR * sizeof( SKP_int ) );
*periodicity_index = k;
}
/* Break early in low-complexity mode if rate distortion is below threshold */
if( lowComplexity && ( rate_dist * 16384.0f < ( SKP_float )SKP_Silk_LTP_gain_middle_avg_RD_Q14 ) ) {
break;
}
}
cbk_ptr_Q14 = SKP_Silk_LTP_vq_ptrs_Q14[ *periodicity_index ];
for( j = 0; j < NB_SUBFR; j++ ) {
SKP_short2float_array( &B[ j * LTP_ORDER ],
&cbk_ptr_Q14[ cbk_index[ j ] * LTP_ORDER ],
LTP_ORDER );
}
for( j = 0; j < NB_SUBFR * LTP_ORDER; j++ ) {
B[ j ] *= Q14_CONVERSION_FAC;
}
}
SKP_int SKP_Silk_encode_frame_FLP(
SKP_Silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */
SKP_uint8 *pCode, /* O Payload */
SKP_int16 *pnBytesOut, /* I/O Number of payload bytes; */
/* input: max length; output: used */
const SKP_int16 *pIn /* I Input speech frame */
)
{
SKP_Silk_encoder_control_FLP sEncCtrl;
SKP_int k, nBytes, ret = 0;
SKP_float *x_frame, *res_pitch_frame;
SKP_int16 pIn_HP[ MAX_FRAME_LENGTH ];
SKP_int16 pIn_HP_LP[ MAX_FRAME_LENGTH ];
SKP_float xfw[ MAX_FRAME_LENGTH ];
SKP_float res_pitch[ 2 * MAX_FRAME_LENGTH + LA_PITCH_MAX ];
SKP_int LBRR_idx, frame_terminator;
/* Low bitrate redundancy parameters */
SKP_uint8 LBRRpayload[ MAX_ARITHM_BYTES ];
SKP_int16 nBytesLBRR;
const SKP_uint16 *FrameTermination_CDF;
sEncCtrl.sCmn.Seed = psEnc->sCmn.frameCounter++ & 3;
/**************************************************************/
/* Setup Input Pointers, and insert frame in input buffer */
/*************************************************************/
/* pointers aligned with start of frame to encode */
x_frame = psEnc->x_buf + psEnc->sCmn.frame_length; // start of frame to encode
res_pitch_frame = res_pitch + psEnc->sCmn.frame_length; // start of pitch LPC residual frame
/****************************/
/* Voice Activity Detection */
/****************************/
SKP_Silk_VAD_FLP( psEnc, &sEncCtrl, pIn );
/*******************************************/
/* High-pass filtering of the input signal */
/*******************************************/
#if HIGH_PASS_INPUT
/* Variable high-pass filter */
SKP_Silk_HP_variable_cutoff_FLP( psEnc, &sEncCtrl, pIn_HP, pIn );
#else
SKP_memcpy( pIn_HP, pIn, psEnc->sCmn.frame_length * sizeof( SKP_int16 ) );
#endif
#if SWITCH_TRANSITION_FILTERING
/* Ensure smooth bandwidth transitions */
SKP_Silk_LP_variable_cutoff( &psEnc->sCmn.sLP, pIn_HP_LP, pIn_HP, psEnc->sCmn.frame_length );
#else
SKP_memcpy( pIn_HP_LP, pIn_HP, psEnc->sCmn.frame_length * sizeof( SKP_int16 ) );
#endif
/*******************************************/
/* Copy new frame to front of input buffer */
/*******************************************/
SKP_short2float_array( x_frame + LA_SHAPE_MS * psEnc->sCmn.fs_kHz, pIn_HP_LP, psEnc->sCmn.frame_length );
/* Add tiny signal to avoid high CPU load from denormalized floating point numbers */
for( k = 0; k < 8; k++ ) {
x_frame[ LA_SHAPE_MS * psEnc->sCmn.fs_kHz + k * ( psEnc->sCmn.frame_length >> 3 ) ] += ( 1 - ( k & 2 ) ) * 1e-6f;
}
/*****************************************/
/* Find pitch lags, initial LPC analysis */
/*****************************************/
SKP_Silk_find_pitch_lags_FLP( psEnc, &sEncCtrl, res_pitch, x_frame );
/************************/
/* Noise shape analysis */
/************************/
SKP_Silk_noise_shape_analysis_FLP( psEnc, &sEncCtrl, res_pitch_frame, x_frame );
/*****************************************/
/* Prefiltering for noise shaper */
/*****************************************/
SKP_Silk_prefilter_FLP( psEnc, &sEncCtrl, xfw, x_frame );
/***************************************************/
/* Find linear prediction coefficients (LPC + LTP) */
/***************************************************/
SKP_Silk_find_pred_coefs_FLP( psEnc, &sEncCtrl, res_pitch );
/****************************************/
/* Process gains */
/****************************************/
SKP_Silk_process_gains_FLP( psEnc, &sEncCtrl );
/****************************************/
/* Low Bitrate Redundant Encoding */
/****************************************/
nBytesLBRR = MAX_ARITHM_BYTES;
SKP_Silk_LBRR_encode_FLP( psEnc, &sEncCtrl, LBRRpayload, &nBytesLBRR, xfw );
/*****************************************/
/* Noise shaping quantization */
/*****************************************/
SKP_Silk_NSQ_wrapper_FLP( psEnc, &sEncCtrl, xfw, psEnc->sCmn.q, 0 );
/**************************************************/
/* Convert speech activity into VAD and DTX flags */
/**************************************************/
if( psEnc->speech_activity < SPEECH_ACTIVITY_DTX_THRES ) {
psEnc->sCmn.vadFlag = NO_VOICE_ACTIVITY;
psEnc->sCmn.noSpeechCounter++;
if( psEnc->sCmn.noSpeechCounter > NO_SPEECH_FRAMES_BEFORE_DTX ) {
psEnc->sCmn.inDTX = 1;
}
if( psEnc->sCmn.noSpeechCounter > MAX_CONSECUTIVE_DTX + NO_SPEECH_FRAMES_BEFORE_DTX ) {
psEnc->sCmn.noSpeechCounter = NO_SPEECH_FRAMES_BEFORE_DTX;
psEnc->sCmn.inDTX = 0;
}
} else {
psEnc->sCmn.noSpeechCounter = 0;
psEnc->sCmn.inDTX = 0;
psEnc->sCmn.vadFlag = VOICE_ACTIVITY;
}
/****************************************/
/* Initialize range coder */
/****************************************/
if( psEnc->sCmn.nFramesInPayloadBuf == 0 ) {
SKP_Silk_range_enc_init( &psEnc->sCmn.sRC );
psEnc->sCmn.nBytesInPayloadBuf = 0;
}
/****************************************/
/* Encode Parameters */
/****************************************/
SKP_Silk_encode_parameters( &psEnc->sCmn, &sEncCtrl.sCmn, &psEnc->sCmn.sRC, psEnc->sCmn.q );
FrameTermination_CDF = SKP_Silk_FrameTermination_CDF;
/****************************************/
/* Update Buffers and State */
/****************************************/
/* Update input buffer */
SKP_memmove( psEnc->x_buf, &psEnc->x_buf[ psEnc->sCmn.frame_length ],
( psEnc->sCmn.frame_length + LA_SHAPE_MS * psEnc->sCmn.fs_kHz ) * sizeof( SKP_float ) );
/* Parameters needed for next frame */
psEnc->sCmn.prev_sigtype = sEncCtrl.sCmn.sigtype;
psEnc->sCmn.prevLag = sEncCtrl.sCmn.pitchL[ NB_SUBFR - 1];
psEnc->sCmn.first_frame_after_reset = 0;
if( psEnc->sCmn.sRC.error ) {
/* Encoder returned error: Clear payload buffer */
psEnc->sCmn.nFramesInPayloadBuf = 0;
} else {
psEnc->sCmn.nFramesInPayloadBuf++;
}
/****************************************/
/* Finalize payload and copy to output */
/****************************************/
if( psEnc->sCmn.nFramesInPayloadBuf * FRAME_LENGTH_MS >= psEnc->sCmn.PacketSize_ms ) {
LBRR_idx = ( psEnc->sCmn.oldest_LBRR_idx + 1 ) & LBRR_IDX_MASK;
/* Check if FEC information should be added */
frame_terminator = SKP_SILK_LAST_FRAME;
if( psEnc->sCmn.LBRR_buffer[ LBRR_idx ].usage == SKP_SILK_ADD_LBRR_TO_PLUS1 ) {
frame_terminator = SKP_SILK_LBRR_VER1;
}
if( psEnc->sCmn.LBRR_buffer[ psEnc->sCmn.oldest_LBRR_idx ].usage == SKP_SILK_ADD_LBRR_TO_PLUS2 ) {
frame_terminator = SKP_SILK_LBRR_VER2;
LBRR_idx = psEnc->sCmn.oldest_LBRR_idx;
}
/* Add the frame termination info to stream */
SKP_Silk_range_encoder( &psEnc->sCmn.sRC, frame_terminator, FrameTermination_CDF );
/* Payload length so far */
SKP_Silk_range_coder_get_length( &psEnc->sCmn.sRC, &nBytes );
/* Check that there is enough space in external output buffer, and move data */
if( *pnBytesOut >= nBytes ) {
SKP_Silk_range_enc_wrap_up( &psEnc->sCmn.sRC );
SKP_memcpy( pCode, psEnc->sCmn.sRC.buffer, nBytes * sizeof( SKP_uint8 ) );
if( frame_terminator > SKP_SILK_MORE_FRAMES &&
*pnBytesOut >= nBytes + psEnc->sCmn.LBRR_buffer[ LBRR_idx ].nBytes ) {
/* Get old packet and add to payload. */
SKP_memcpy( &pCode[ nBytes ],
psEnc->sCmn.LBRR_buffer[ LBRR_idx ].payload,
psEnc->sCmn.LBRR_buffer[ LBRR_idx ].nBytes * sizeof( SKP_uint8 ) );
nBytes += psEnc->sCmn.LBRR_buffer[ LBRR_idx ].nBytes;
}
*pnBytesOut = nBytes;
/* Update FEC buffer */
SKP_memcpy( psEnc->sCmn.LBRR_buffer[ psEnc->sCmn.oldest_LBRR_idx ].payload, LBRRpayload,
nBytesLBRR * sizeof( SKP_uint8 ) );
psEnc->sCmn.LBRR_buffer[ psEnc->sCmn.oldest_LBRR_idx ].nBytes = nBytesLBRR;
/* The line below describes how FEC should be used */
psEnc->sCmn.LBRR_buffer[ psEnc->sCmn.oldest_LBRR_idx ].usage = sEncCtrl.sCmn.LBRR_usage;
psEnc->sCmn.oldest_LBRR_idx = ( ( psEnc->sCmn.oldest_LBRR_idx + 1 ) & LBRR_IDX_MASK );
} else {
/* Not enough space: Payload will be discarded */
*pnBytesOut = 0;
nBytes = 0;
ret = SKP_SILK_ENC_PAYLOAD_BUF_TOO_SHORT;
}
/* Reset the number of frames in payload buffer */
psEnc->sCmn.nFramesInPayloadBuf = 0;
} else {
/* No payload this time */
*pnBytesOut = 0;
/* Encode that more frames follows */
frame_terminator = SKP_SILK_MORE_FRAMES;
SKP_Silk_range_encoder( &psEnc->sCmn.sRC, frame_terminator, FrameTermination_CDF );
/* Payload length so far */
SKP_Silk_range_coder_get_length( &psEnc->sCmn.sRC, &nBytes );
}
/* Check for arithmetic coder errors */
if( psEnc->sCmn.sRC.error ) {
ret = SKP_SILK_ENC_INTERNAL_ERROR;
}
/* Simulate number of ms buffered in channel because of exceeding TargetRate */
psEnc->BufferedInChannel_ms += ( 8.0f * 1000.0f * ( nBytes - psEnc->sCmn.nBytesInPayloadBuf ) ) / psEnc->sCmn.TargetRate_bps;
psEnc->BufferedInChannel_ms -= FRAME_LENGTH_MS;
psEnc->BufferedInChannel_ms = SKP_LIMIT_float( psEnc->BufferedInChannel_ms, 0.0f, 100.0f );
psEnc->sCmn.nBytesInPayloadBuf = nBytes;
if( psEnc->speech_activity > WB_DETECT_ACTIVE_SPEECH_LEVEL_THRES ) {
psEnc->sCmn.sSWBdetect.ActiveSpeech_ms = SKP_ADD_POS_SAT32( psEnc->sCmn.sSWBdetect.ActiveSpeech_ms, FRAME_LENGTH_MS );
}
return( ret );
}
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 );
} 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 ];
#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 = SKP_LSHIFT( psEnc->sCmn.frame_length, 1 ) + LA_SHAPE_MS * psEnc->sCmn.fs_kHz;
#ifndef FIXED_POINT
SKP_float2short_array( x_bufFIX, psEnc->x_buf, nSamples_temp );
#endif
if( SKP_SMULBB( fs_kHz, 1000 ) < psEnc->sCmn.API_fs_Hz && psEnc->sCmn.fs_kHz != 0 ) {
/* Resample buffered data in x_buf to API_fs_Hz */
silk_resampler_state_struct temp_resampler_state;
/* Initialize resampler for temporary resampling of x_buf data to API_fs_Hz */
ret += silk_resampler_init( &temp_resampler_state, SKP_SMULBB( psEnc->sCmn.fs_kHz, 1000 ), psEnc->sCmn.API_fs_Hz );
/* 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 = SKP_DIV32_16( nSamples_temp * psEnc->sCmn.API_fs_Hz, SKP_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, SKP_SMULBB( fs_kHz, 1000 ) );
} else {
/* Copy data */
SKP_memcpy( x_buf_API_fs_Hz, x_bufFIX, nSamples_temp * sizeof( opus_int16 ) );
}
if( 1000 * fs_kHz != psEnc->sCmn.API_fs_Hz ) {
/* Correct resampler state (unless resampling by a factor 1) 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
SKP_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;
}
