cannam@154: /*********************************************************************** cannam@154: Copyright (c) 2006-2011, Skype Limited. All rights reserved. cannam@154: Redistribution and use in source and binary forms, with or without cannam@154: modification, are permitted provided that the following conditions cannam@154: are met: cannam@154: - Redistributions of source code must retain the above copyright notice, cannam@154: this list of conditions and the following disclaimer. cannam@154: - Redistributions in binary form must reproduce the above copyright cannam@154: notice, this list of conditions and the following disclaimer in the cannam@154: documentation and/or other materials provided with the distribution. cannam@154: - Neither the name of Internet Society, IETF or IETF Trust, nor the cannam@154: names of specific contributors, may be used to endorse or promote cannam@154: products derived from this software without specific prior written cannam@154: permission. cannam@154: THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" cannam@154: AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE cannam@154: IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE cannam@154: ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE cannam@154: LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR cannam@154: CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF cannam@154: SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS cannam@154: INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN cannam@154: CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) cannam@154: ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE cannam@154: POSSIBILITY OF SUCH DAMAGE. cannam@154: ***********************************************************************/ cannam@154: cannam@154: #ifdef HAVE_CONFIG_H cannam@154: #include "config.h" cannam@154: #endif cannam@154: cannam@154: #include cannam@154: #include "main_FLP.h" cannam@154: #include "tuning_parameters.h" cannam@154: cannam@154: void silk_find_pitch_lags_FLP( cannam@154: silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ cannam@154: silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ cannam@154: silk_float res[], /* O Residual */ cannam@154: const silk_float x[], /* I Speech signal */ cannam@154: int arch /* I Run-time architecture */ cannam@154: ) cannam@154: { cannam@154: opus_int buf_len; cannam@154: silk_float thrhld, res_nrg; cannam@154: const silk_float *x_buf_ptr, *x_buf; cannam@154: silk_float auto_corr[ MAX_FIND_PITCH_LPC_ORDER + 1 ]; cannam@154: silk_float A[ MAX_FIND_PITCH_LPC_ORDER ]; cannam@154: silk_float refl_coef[ MAX_FIND_PITCH_LPC_ORDER ]; cannam@154: silk_float Wsig[ FIND_PITCH_LPC_WIN_MAX ]; cannam@154: silk_float *Wsig_ptr; cannam@154: cannam@154: /******************************************/ cannam@154: /* Set up buffer lengths etc based on Fs */ cannam@154: /******************************************/ cannam@154: buf_len = psEnc->sCmn.la_pitch + psEnc->sCmn.frame_length + psEnc->sCmn.ltp_mem_length; cannam@154: cannam@154: /* Safety check */ cannam@154: celt_assert( buf_len >= psEnc->sCmn.pitch_LPC_win_length ); cannam@154: cannam@154: x_buf = x - psEnc->sCmn.ltp_mem_length; cannam@154: cannam@154: /******************************************/ cannam@154: /* Estimate LPC AR coeficients */ cannam@154: /******************************************/ cannam@154: cannam@154: /* Calculate windowed signal */ cannam@154: cannam@154: /* First LA_LTP samples */ cannam@154: x_buf_ptr = x_buf + buf_len - psEnc->sCmn.pitch_LPC_win_length; cannam@154: Wsig_ptr = Wsig; cannam@154: silk_apply_sine_window_FLP( Wsig_ptr, x_buf_ptr, 1, psEnc->sCmn.la_pitch ); cannam@154: cannam@154: /* Middle non-windowed samples */ cannam@154: Wsig_ptr += psEnc->sCmn.la_pitch; cannam@154: x_buf_ptr += psEnc->sCmn.la_pitch; cannam@154: silk_memcpy( Wsig_ptr, x_buf_ptr, ( psEnc->sCmn.pitch_LPC_win_length - ( psEnc->sCmn.la_pitch << 1 ) ) * sizeof( silk_float ) ); cannam@154: cannam@154: /* Last LA_LTP samples */ cannam@154: Wsig_ptr += psEnc->sCmn.pitch_LPC_win_length - ( psEnc->sCmn.la_pitch << 1 ); cannam@154: x_buf_ptr += psEnc->sCmn.pitch_LPC_win_length - ( psEnc->sCmn.la_pitch << 1 ); cannam@154: silk_apply_sine_window_FLP( Wsig_ptr, x_buf_ptr, 2, psEnc->sCmn.la_pitch ); cannam@154: cannam@154: /* Calculate autocorrelation sequence */ cannam@154: silk_autocorrelation_FLP( auto_corr, Wsig, psEnc->sCmn.pitch_LPC_win_length, psEnc->sCmn.pitchEstimationLPCOrder + 1 ); cannam@154: cannam@154: /* Add white noise, as a fraction of the energy */ cannam@154: auto_corr[ 0 ] += auto_corr[ 0 ] * FIND_PITCH_WHITE_NOISE_FRACTION + 1; cannam@154: cannam@154: /* Calculate the reflection coefficients using Schur */ cannam@154: res_nrg = silk_schur_FLP( refl_coef, auto_corr, psEnc->sCmn.pitchEstimationLPCOrder ); cannam@154: cannam@154: /* Prediction gain */ cannam@154: psEncCtrl->predGain = auto_corr[ 0 ] / silk_max_float( res_nrg, 1.0f ); cannam@154: cannam@154: /* Convert reflection coefficients to prediction coefficients */ cannam@154: silk_k2a_FLP( A, refl_coef, psEnc->sCmn.pitchEstimationLPCOrder ); cannam@154: cannam@154: /* Bandwidth expansion */ cannam@154: silk_bwexpander_FLP( A, psEnc->sCmn.pitchEstimationLPCOrder, FIND_PITCH_BANDWIDTH_EXPANSION ); cannam@154: cannam@154: /*****************************************/ cannam@154: /* LPC analysis filtering */ cannam@154: /*****************************************/ cannam@154: silk_LPC_analysis_filter_FLP( res, A, x_buf, buf_len, psEnc->sCmn.pitchEstimationLPCOrder ); cannam@154: cannam@154: if( psEnc->sCmn.indices.signalType != TYPE_NO_VOICE_ACTIVITY && psEnc->sCmn.first_frame_after_reset == 0 ) { cannam@154: /* Threshold for pitch estimator */ cannam@154: thrhld = 0.6f; cannam@154: thrhld -= 0.004f * psEnc->sCmn.pitchEstimationLPCOrder; cannam@154: thrhld -= 0.1f * psEnc->sCmn.speech_activity_Q8 * ( 1.0f / 256.0f ); cannam@154: thrhld -= 0.15f * (psEnc->sCmn.prevSignalType >> 1); cannam@154: thrhld -= 0.1f * psEnc->sCmn.input_tilt_Q15 * ( 1.0f / 32768.0f ); cannam@154: cannam@154: /*****************************************/ cannam@154: /* Call Pitch estimator */ cannam@154: /*****************************************/ cannam@154: if( silk_pitch_analysis_core_FLP( res, psEncCtrl->pitchL, &psEnc->sCmn.indices.lagIndex, cannam@154: &psEnc->sCmn.indices.contourIndex, &psEnc->LTPCorr, psEnc->sCmn.prevLag, psEnc->sCmn.pitchEstimationThreshold_Q16 / 65536.0f, cannam@154: thrhld, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr, arch ) == 0 ) cannam@154: { cannam@154: psEnc->sCmn.indices.signalType = TYPE_VOICED; cannam@154: } else { cannam@154: psEnc->sCmn.indices.signalType = TYPE_UNVOICED; cannam@154: } cannam@154: } else { cannam@154: silk_memset( psEncCtrl->pitchL, 0, sizeof( psEncCtrl->pitchL ) ); cannam@154: psEnc->sCmn.indices.lagIndex = 0; cannam@154: psEnc->sCmn.indices.contourIndex = 0; cannam@154: psEnc->LTPCorr = 0; cannam@154: } cannam@154: }