fe_noise.cpp

这是一个语音特征提取的程序源码

///////////////////////////////////////////////////////////////////////////////
// This is a part of the Feature program.
// Version: 1.0
// Date: February 22, 2003
// Programmer: Oh-Wook Kwon
// Copyright(c) 2003 Oh-Wook Kwon. All rights reserved. [e-mail]
///////////////////////////////////////////////////////////////////////////////

#include "StdAfx.h"
#include "FE_feature.h"

static const long A=16807L;
static const long M=2147483647L;

static long g_noise_In[16] = { 0L,
1973272912L, 747177549L, 20464843L,
640830765L, 1098742207L, 78126602L,
84743774L, 831312807L, 124667236L,
1172177002L, 1124933064L, 1223960546L,
1878892440L, 1449793615L, 553303732L
};

static int g_noise_strm	= 2;
static double RandomLE();
static double RandomBE();


double Fe::Random()
{
	if(m_byteOrder==MY_LITTLE_ENDIAN) return RandomLE();
	else return RandomBE();
}


double RandomLE()
{
	short 	*p,*q,k;
	long 	Hi,Lo;
	
	p = (short *)&g_noise_In[g_noise_strm];
	Hi = *(p+1) * A;
	*(p+1) = 0;
	Lo = g_noise_In[g_noise_strm]*A;
	p = (short *)&Lo;
	Hi += *(p+1);
	q = (short *)&Hi;
	*(p+1) = *q&0x7fff;
	k = *(q+1)<<1;
	if(*q&0x8000) k++;
	Lo -= M;
	Lo += k;
	if(Lo<0) Lo+=M;
	g_noise_In[g_noise_strm] = Lo;
	return((double)Lo*4.656612875E-10);
}


double 	RandomBE()
{
	short 	*p,*q,k;
	long 	Hi,Lo;
	
	p = (short *)&g_noise_In[g_noise_strm];
	Hi = *(p)*A;
	*(p) = 0;
	Lo = g_noise_In[g_noise_strm]*A;
	p = (short *)&Lo;
	Hi += *(p);
	q = (short *)&Hi;
	*(p) = *(q+1)&0x7fff;
	k = *(q)<<1;
	if(*(q+1)&0x8000) k++;
	Lo -= M;
	Lo += k;
	if(Lo<0) Lo += M;
	g_noise_In[g_noise_strm] = Lo;
	return((double)Lo*4.656612875E-10);
}


double Fe::GaussianNoise(double x, double s)
{
	double 	v1,v2,w,z1;
	static 	double 	z2=0.0;
	
	if (z2 != 0.0) {
		z1=z2;
		z2=0.0;
	}
	else {
		do {
			v1 = 2.0*Random()-1.0;
			v2 = 2.0*Random()-1.0;
			w = v1*v1 + v2*v2;
		} while (w >= 1.0);
		w = sqrt((-2.0*log(w))/w);
		z1 = v1*w;
		z2 = v2*w;
	}
	return(x+z1*s);
}


/********************************************************************** 
 *	add additive white Gaussian noise(AWGN) into speech signal
 **********************************************************************/
int Fe::AddNoise(float *sample, int num_of_samples, float *waveform, int insert_noise, float SNR)
{
	int	j;
	float	Ps, Pn;
	if (insert_noise) {
		Ps = 0.0;
		for (j=0; j<num_of_samples; j++)
			Ps += sample[j]*sample[j];
		Ps /= (float) num_of_samples;
		Pn = Ps / EXP10( 0.1 * SNR );
		for (j=0; j<num_of_samples; j++)
			waveform[j] = sample[j] + (float) GaussianNoise( 0.0, sqrt((double) Pn) );
	}
	return(1);
}


/* ------------------------------------------------------------------ 
When using ulaw data, often entire frames will contain only zeros.
This is bad for spectral calculation, so we should add a dither
signal to make sure the signal is always at least some quantization noise.

 12-Oct-94
 Now using rand48 routines instead for hpux compatibility
------------------------------------------------------------------ */
/* add 1-bit dither to signal */
int Fe::Dither(float *buf, int n)
{
	int 	i;
	srand(1);
	for (i=0;i<n;i++) buf[i] += ((rand()%2)?1:0);
	return(1);
	
}