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 );
}
/* High-pass filter with cutoff frequency adaptation based on pitch lag statistics */
void SKP_Silk_HP_variable_cutoff_FLP(
SKP_Silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */
SKP_Silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */
SKP_int16 *out, /* O High-pass filtered output signal */
const SKP_int16 *in /* I Input signal */
)
{
SKP_float pitch_freq_Hz, pitch_freq_log, quality, delta_freq, smth_coef, Fc, r;
SKP_int32 B_Q28[ 3 ], A_Q28[ 2 ];
/*********************************************/
/* Estimate low end of pitch frequency range */
/*********************************************/
if( psEnc->sCmn.prev_sigtype == SIG_TYPE_VOICED ) {
/* Difference, in log domain */
pitch_freq_Hz = 1e3f * psEnc->sCmn.fs_kHz / psEnc->sCmn.prevLag;
pitch_freq_log = SKP_Silk_log2( pitch_freq_Hz );
/* Adjustment based on quality */
quality = psEncCtrl->input_quality_bands[ 0 ];
pitch_freq_log -= quality * quality * ( pitch_freq_log - SKP_Silk_log2( VARIABLE_HP_MIN_FREQ ) );
pitch_freq_log += 0.5f * ( 0.6f - quality );
delta_freq = pitch_freq_log - psEnc->variable_HP_smth1;
if( delta_freq < 0.0 ) {
/* Less smoothing for decreasing pitch frequency, to track something close to the minimum */
delta_freq *= 3.0f;
}
/* Limit delta, to reduce impact of outliers */
delta_freq = SKP_LIMIT_float( delta_freq, -VARIABLE_HP_MAX_DELTA_FREQ, VARIABLE_HP_MAX_DELTA_FREQ );
/* Update smoother */
smth_coef = VARIABLE_HP_SMTH_COEF1 * psEnc->speech_activity;
psEnc->variable_HP_smth1 += smth_coef * delta_freq;
}
/* Second smoother */
psEnc->variable_HP_smth2 += VARIABLE_HP_SMTH_COEF2 * ( psEnc->variable_HP_smth1 - psEnc->variable_HP_smth2 );
/* Convert from log scale to Hertz */
psEncCtrl->pitch_freq_low_Hz = ( SKP_float )pow( 2.0f, psEnc->variable_HP_smth2 );
/* Limit frequency range */
psEncCtrl->pitch_freq_low_Hz = SKP_LIMIT_float( psEncCtrl->pitch_freq_low_Hz, VARIABLE_HP_MIN_FREQ, VARIABLE_HP_MAX_FREQ );
/*******************************/
/* Compute filter coefficients */
/*******************************/
/* Compute cut-off frequency, in radians */
Fc = ( SKP_float )( 0.45f * 2.0f * 3.14159265359 * psEncCtrl->pitch_freq_low_Hz / ( 1e3f * psEnc->sCmn.fs_kHz ) );
/* 2nd order ARMA coefficients */
r = 1.0f - 0.92f * Fc;
/* b = r * [1; -2; 1]; */
/* a = [1; -2 * r * (1 - 0.5 * Fc^2); r^2]; */
B_Q28[ 0 ] = SKP_float2int( ( 1 << 28 ) * r );
B_Q28[ 1 ] = SKP_float2int( ( 1 << 28 ) * -2.0f * r );
B_Q28[ 2 ] = B_Q28[ 0 ];
A_Q28[ 0 ] = SKP_float2int( ( 1 << 28 ) * -2.0f * r * ( 1.0f - 0.5f * Fc * Fc ) );
A_Q28[ 1 ] = SKP_float2int( ( 1 << 28 ) * r * r );
/********************/
/* High-pass filter */
/********************/
SKP_Silk_biquad_alt( in, B_Q28, A_Q28, psEnc->sCmn.In_HP_State, out, psEnc->sCmn.frame_length );
}