Files
LuaCsForBarotraumaEP/Libraries/Concentus/CSharp/Concentus/Silk/VoiceActivityDetection.cs

410 lines
17 KiB
C#

/* Copyright (c) 2006-2011 Skype Limited. All Rights Reserved
Ported to C# by Logan Stromberg
Redistribution and use in source and binary forms, with or without
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 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.
*/
namespace Concentus.Silk
{
using Concentus.Common;
using Concentus.Common.CPlusPlus;
using Concentus.Silk.Enums;
using Concentus.Silk.Structs;
using System.Diagnostics;
/// <summary>
/// Voice Activity Detection module for silk codec
/// </summary>
internal static class VoiceActivityDetection
{
/// <summary>
/// Weighting factors for tilt measure
/// </summary>
private static readonly int[] tiltWeights = { 30000, 6000, -12000, -12000 };
/// <summary>
/// Initialization of the Silk VAD
/// </summary>
/// <param name="psSilk_VAD">O Pointer to Silk VAD state. Cannot be nullptr</param>
/// <returns>0 if success</returns>
internal static int silk_VAD_Init(SilkVADState psSilk_VAD)
{
int b, ret = 0;
/* reset state memory */
psSilk_VAD.Reset();
/* init noise levels */
/* Initialize array with approx pink noise levels (psd proportional to inverse of frequency) */
for (b = 0; b < SilkConstants.VAD_N_BANDS; b++)
{
psSilk_VAD.NoiseLevelBias[b] = Inlines.silk_max_32(Inlines.silk_DIV32_16(SilkConstants.VAD_NOISE_LEVELS_BIAS, (short)(b + 1)), 1);
}
/* Initialize state */
for (b = 0; b < SilkConstants.VAD_N_BANDS; b++)
{
psSilk_VAD.NL[b] = Inlines.silk_MUL(100, psSilk_VAD.NoiseLevelBias[b]);
psSilk_VAD.inv_NL[b] = Inlines.silk_DIV32(int.MaxValue, psSilk_VAD.NL[b]);
}
psSilk_VAD.counter = 15;
/* init smoothed energy-to-noise ratio*/
for (b = 0; b < SilkConstants.VAD_N_BANDS; b++)
{
psSilk_VAD.NrgRatioSmth_Q8[b] = 100 * 256; /* 100 * 256 -. 20 dB SNR */
}
return (ret);
}
/// <summary>
/// Get the speech activity level in Q8
/// </summary>
/// <param name="psEncC">I/O Encoder state</param>
/// <param name="pIn">I PCM input</param>
/// <returns>0 if success</returns>
internal static int silk_VAD_GetSA_Q8(
SilkChannelEncoder psEncC,
short[] pIn,
int pIn_ptr)
{
int SA_Q15, pSNR_dB_Q7, input_tilt;
int decimated_framelength1, decimated_framelength2;
int decimated_framelength;
int dec_subframe_length, dec_subframe_offset, SNR_Q7, i, b, s;
int sumSquared = 0, smooth_coef_Q16;
short HPstateTmp;
short[] X;
int[] Xnrg = new int[SilkConstants.VAD_N_BANDS];
int[] NrgToNoiseRatio_Q8 = new int[SilkConstants.VAD_N_BANDS];
int speech_nrg, x_tmp;
int[] X_offset = new int[SilkConstants.VAD_N_BANDS];
int ret = 0;
SilkVADState psSilk_VAD = psEncC.sVAD;
/* Safety checks */
Inlines.OpusAssert(SilkConstants.VAD_N_BANDS == 4);
Inlines.OpusAssert(SilkConstants.MAX_FRAME_LENGTH >= psEncC.frame_length);
Inlines.OpusAssert(psEncC.frame_length <= 512);
Inlines.OpusAssert(psEncC.frame_length == 8 * Inlines.silk_RSHIFT(psEncC.frame_length, 3));
/***********************/
/* Filter and Decimate */
/***********************/
decimated_framelength1 = Inlines.silk_RSHIFT(psEncC.frame_length, 1);
decimated_framelength2 = Inlines.silk_RSHIFT(psEncC.frame_length, 2);
decimated_framelength = Inlines.silk_RSHIFT(psEncC.frame_length, 3);
/* Decimate into 4 bands:
0 L 3L L 3L 5L
- -- - -- --
8 8 2 4 4
[0-1 kHz| temp. |1-2 kHz| 2-4 kHz | 4-8 kHz |
They're arranged to allow the minimal ( frame_length / 4 ) extra
scratch space during the downsampling process */
X_offset[0] = 0;
X_offset[1] = decimated_framelength + decimated_framelength2;
X_offset[2] = X_offset[1] + decimated_framelength;
X_offset[3] = X_offset[2] + decimated_framelength2;
X = new short[X_offset[3] + decimated_framelength1];
/* 0-8 kHz to 0-4 kHz and 4-8 kHz */
Filters.silk_ana_filt_bank_1(pIn, pIn_ptr, psSilk_VAD.AnaState,
X, X, X_offset[3], psEncC.frame_length);
/* 0-4 kHz to 0-2 kHz and 2-4 kHz */
Filters.silk_ana_filt_bank_1(X, 0, psSilk_VAD.AnaState1,
X, X, X_offset[2], decimated_framelength1);
/* 0-2 kHz to 0-1 kHz and 1-2 kHz */
Filters.silk_ana_filt_bank_1(X, 0, psSilk_VAD.AnaState2,
X, X, X_offset[1], decimated_framelength2);
/*********************************************/
/* HP filter on lowest band (differentiator) */
/*********************************************/
X[decimated_framelength - 1] = (short)(Inlines.silk_RSHIFT(X[decimated_framelength - 1], 1));
HPstateTmp = X[decimated_framelength - 1];
for (i = decimated_framelength - 1; i > 0; i--)
{
X[i - 1] = (short)(Inlines.silk_RSHIFT(X[i - 1], 1));
X[i] -= X[i - 1];
}
X[0] -= psSilk_VAD.HPstate;
psSilk_VAD.HPstate = HPstateTmp;
/*************************************/
/* Calculate the energy in each band */
/*************************************/
for (b = 0; b < SilkConstants.VAD_N_BANDS; b++)
{
/* Find the decimated framelength in the non-uniformly divided bands */
decimated_framelength = Inlines.silk_RSHIFT(psEncC.frame_length, Inlines.silk_min_int(SilkConstants.VAD_N_BANDS - b, SilkConstants.VAD_N_BANDS - 1));
/* Split length into subframe lengths */
dec_subframe_length = Inlines.silk_RSHIFT(decimated_framelength, SilkConstants.VAD_INTERNAL_SUBFRAMES_LOG2);
dec_subframe_offset = 0;
/* Compute energy per sub-frame */
/* initialize with summed energy of last subframe */
Xnrg[b] = psSilk_VAD.XnrgSubfr[b];
for (s = 0; s < SilkConstants.VAD_INTERNAL_SUBFRAMES; s++)
{
sumSquared = 0;
for (i = 0; i < dec_subframe_length; i++)
{
/* The energy will be less than dec_subframe_length * ( silk_int16_MIN / 8 ) ^ 2. */
/* Therefore we can accumulate with no risk of overflow (unless dec_subframe_length > 128) */
x_tmp = Inlines.silk_RSHIFT(
X[X_offset[b] + i + dec_subframe_offset], 3);
sumSquared = Inlines.silk_SMLABB(sumSquared, x_tmp, x_tmp);
/* Safety check */
Inlines.OpusAssert(sumSquared >= 0);
}
/* Add/saturate summed energy of current subframe */
if (s < SilkConstants.VAD_INTERNAL_SUBFRAMES - 1)
{
Xnrg[b] = Inlines.silk_ADD_POS_SAT32(Xnrg[b], sumSquared);
}
else
{
/* Look-ahead subframe */
Xnrg[b] = Inlines.silk_ADD_POS_SAT32(Xnrg[b], Inlines.silk_RSHIFT(sumSquared, 1));
}
dec_subframe_offset += dec_subframe_length;
}
psSilk_VAD.XnrgSubfr[b] = sumSquared;
}
/********************/
/* Noise estimation */
/********************/
silk_VAD_GetNoiseLevels(Xnrg, psSilk_VAD);
/***********************************************/
/* Signal-plus-noise to noise ratio estimation */
/***********************************************/
sumSquared = 0;
input_tilt = 0;
for (b = 0; b < SilkConstants.VAD_N_BANDS; b++)
{
speech_nrg = Xnrg[b] - psSilk_VAD.NL[b];
if (speech_nrg > 0)
{
/* Divide, with sufficient resolution */
if ((Xnrg[b] & 0xFF800000) == 0)
{
NrgToNoiseRatio_Q8[b] = Inlines.silk_DIV32(Inlines.silk_LSHIFT(Xnrg[b], 8), psSilk_VAD.NL[b] + 1);
}
else {
NrgToNoiseRatio_Q8[b] = Inlines.silk_DIV32(Xnrg[b], Inlines.silk_RSHIFT(psSilk_VAD.NL[b], 8) + 1);
}
/* Convert to log domain */
SNR_Q7 = Inlines.silk_lin2log(NrgToNoiseRatio_Q8[b]) - 8 * 128;
/* Sum-of-squares */
sumSquared = Inlines.silk_SMLABB(sumSquared, SNR_Q7, SNR_Q7); /* Q14 */
/* Tilt measure */
if (speech_nrg < ((int)1 << 20))
{
/* Scale down SNR value for small subband speech energies */
SNR_Q7 = Inlines.silk_SMULWB(Inlines.silk_LSHIFT(Inlines.silk_SQRT_APPROX(speech_nrg), 6), SNR_Q7);
}
input_tilt = Inlines.silk_SMLAWB(input_tilt, tiltWeights[b], SNR_Q7);
}
else
{
NrgToNoiseRatio_Q8[b] = 256;
}
}
/* Mean-of-squares */
sumSquared = Inlines.silk_DIV32_16(sumSquared, SilkConstants.VAD_N_BANDS); /* Q14 */
/* Root-mean-square approximation, scale to dBs, and write to output pointer */
pSNR_dB_Q7 = (short)(3 * Inlines.silk_SQRT_APPROX(sumSquared)); /* Q7 */
/*********************************/
/* Speech Probability Estimation */
/*********************************/
SA_Q15 = Sigmoid.silk_sigm_Q15(Inlines.silk_SMULWB(SilkConstants.VAD_SNR_FACTOR_Q16, pSNR_dB_Q7) - SilkConstants.VAD_NEGATIVE_OFFSET_Q5);
/**************************/
/* Frequency Tilt Measure */
/**************************/
psEncC.input_tilt_Q15 = Inlines.silk_LSHIFT(Sigmoid.silk_sigm_Q15(input_tilt) - 16384, 1);
/**************************************************/
/* Scale the sigmoid output based on power levels */
/**************************************************/
speech_nrg = 0;
for (b = 0; b < SilkConstants.VAD_N_BANDS; b++)
{
/* Accumulate signal-without-noise energies, higher frequency bands have more weight */
speech_nrg += (b + 1) * Inlines.silk_RSHIFT(Xnrg[b] - psSilk_VAD.NL[b], 4);
}
/* Power scaling */
if (speech_nrg <= 0)
{
SA_Q15 = Inlines.silk_RSHIFT(SA_Q15, 1);
}
else if (speech_nrg < 32768)
{
if (psEncC.frame_length == 10 * psEncC.fs_kHz)
{
speech_nrg = Inlines.silk_LSHIFT_SAT32(speech_nrg, 16);
}
else
{
speech_nrg = Inlines.silk_LSHIFT_SAT32(speech_nrg, 15);
}
/* square-root */
speech_nrg = Inlines.silk_SQRT_APPROX(speech_nrg);
SA_Q15 = Inlines.silk_SMULWB(32768 + speech_nrg, SA_Q15);
}
/* Copy the resulting speech activity in Q8 */
psEncC.speech_activity_Q8 = Inlines.silk_min_int(Inlines.silk_RSHIFT(SA_Q15, 7), byte.MaxValue);
/***********************************/
/* Energy Level and SNR estimation */
/***********************************/
/* Smoothing coefficient */
smooth_coef_Q16 = Inlines.silk_SMULWB(SilkConstants.VAD_SNR_SMOOTH_COEF_Q18, Inlines.silk_SMULWB((int)SA_Q15, SA_Q15));
if (psEncC.frame_length == 10 * psEncC.fs_kHz)
{
smooth_coef_Q16 >>= 1;
}
for (b = 0; b < SilkConstants.VAD_N_BANDS; b++)
{
/* compute smoothed energy-to-noise ratio per band */
psSilk_VAD.NrgRatioSmth_Q8[b] = Inlines.silk_SMLAWB(psSilk_VAD.NrgRatioSmth_Q8[b],
NrgToNoiseRatio_Q8[b] - psSilk_VAD.NrgRatioSmth_Q8[b], smooth_coef_Q16);
/* signal to noise ratio in dB per band */
SNR_Q7 = 3 * (Inlines.silk_lin2log(psSilk_VAD.NrgRatioSmth_Q8[b]) - 8 * 128);
/* quality = sigmoid( 0.25 * ( SNR_dB - 16 ) ); */
psEncC.input_quality_bands_Q15[b] = Sigmoid.silk_sigm_Q15(Inlines.silk_RSHIFT(SNR_Q7 - 16 * 128, 4));
}
return (ret);
}
/// <summary>
/// Noise level estimation
/// </summary>
/// <param name="pX">I subband energies [VAD_N_BANDS]</param>
/// <param name="psSilk_VAD">I/O Pointer to Silk VAD state</param>
internal static void silk_VAD_GetNoiseLevels(
int[] pX,
SilkVADState psSilk_VAD)
{
int k;
int nl, nrg, inv_nrg;
int coef, min_coef;
/* Initially faster smoothing */
if (psSilk_VAD.counter < 1000)
{ /* 1000 = 20 sec */
min_coef = Inlines.silk_DIV32_16(short.MaxValue, (short)(Inlines.silk_RSHIFT(psSilk_VAD.counter, 4) + 1));
}
else
{
min_coef = 0;
}
for (k = 0; k < SilkConstants.VAD_N_BANDS; k++)
{
/* Get old noise level estimate for current band */
nl = psSilk_VAD.NL[k];
Inlines.OpusAssert(nl >= 0);
/* Add bias */
nrg = Inlines.silk_ADD_POS_SAT32(pX[k], psSilk_VAD.NoiseLevelBias[k]);
Inlines.OpusAssert(nrg > 0);
/* Invert energies */
inv_nrg = Inlines.silk_DIV32(int.MaxValue, nrg);
Inlines.OpusAssert(inv_nrg >= 0);
/* Less update when subband energy is high */
if (nrg > Inlines.silk_LSHIFT(nl, 3))
{
coef = SilkConstants.VAD_NOISE_LEVEL_SMOOTH_COEF_Q16 >> 3;
}
else if (nrg < nl)
{
coef = SilkConstants.VAD_NOISE_LEVEL_SMOOTH_COEF_Q16;
}
else
{
coef = Inlines.silk_SMULWB(Inlines.silk_SMULWW(inv_nrg, nl), SilkConstants.VAD_NOISE_LEVEL_SMOOTH_COEF_Q16 << 1);
}
/* Initially faster smoothing */
coef = Inlines.silk_max_int(coef, min_coef);
/* Smooth inverse energies */
psSilk_VAD.inv_NL[k] = Inlines.silk_SMLAWB(psSilk_VAD.inv_NL[k], inv_nrg - psSilk_VAD.inv_NL[k], coef);
Inlines.OpusAssert(psSilk_VAD.inv_NL[k] >= 0);
/* Compute noise level by inverting again */
nl = Inlines.silk_DIV32(int.MaxValue, psSilk_VAD.inv_NL[k]);
Inlines.OpusAssert(nl >= 0);
/* Limit noise levels (guarantee 7 bits of head room) */
nl = Inlines.silk_min(nl, 0x00FFFFFF);
/* Store as part of state */
psSilk_VAD.NL[k] = nl;
}
/* Increment frame counter */
psSilk_VAD.counter++;
}
}
}