home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
AmigActive 10
/
AACD 10.iso
/
AACD
/
Online
/
SpeakFreely
/
src
/
lpc10
/
voice.c
< prev
next >
Wrap
C/C++ Source or Header
|
2000-05-18
|
11KB
|
360 lines
/***************************************************************************
*
* VOICIN Version 52
*
***************************************************************************
*
* Voicing Detection (VOICIN) makes voicing decisions for each half
* frame of input speech. Tentative voicing decisions are made two frames
* in the future (2F) for each half frame. These decisions are carried
* through one frame in the future (1F) to the present (P) frame where
* they are examined and smoothed, resulting in the final voicing
* decisions for each half frame.
* The voicing parameter (signal measurement) column vector (VALUE)
* is based on a rectangular window of speech samples determined by the
* window placement algorithm. The voicing parameter vector contains the
* AMDF windowed maximum-to-minimum ratio, the zero crossing rate, energy
* measures, reflection coefficients, and prediction gains. The voicing
* window is placed to avoid contamination of the voicing parameter vector
* with speech onsets.
* The input signal is then classified as unvoiced (including
* silence) or voiced. This decision is made by a linear discriminant
* function consisting of a dot product of the voicing decision
* coefficient (VDC) row vector with the measurement column vector
* (VALUE). The VDC vector is 2-dimensional, each row vector is optimized
* for a particular signal-to-noise ratio (SNR). So, before the dot
* product is performed, the SNR is estimated to select the appropriate
* VDC vector.
* The smoothing algorithm is a modified median smoother. The
* voicing discriminant function is used by the smoother to determine how
* strongly voiced or unvoiced a signal is. The smoothing is further
* modified if a speech onset and a voicing decision transition occur
* within one half frame. In this case, the voicing decision transition
* is extended to the speech onset. For transmission purposes, there are
* constraints on the duration and transition of voicing decisions. The
* smoother takes these constraints into account.
* Finally, the energy estimates are updated along with the dither
* threshold used to calculate the zero crossing rate (ZC).
*
* Inputs:
* VWIN - Voicing window limits
* INBUF - Input speech buffer
* LPBUF - Low-pass filtered speech buffer
* BUFLIM - INBUF and LPBUF limits
* HALF - Present analysis half frame number
* MINAMD - Minimum value of the AMDF
* MAXAMD - Maximum value of the AMDF
* MINTAU - Pointer to the lag of the minimum AMDF value
* IVRC(2) - Inverse filter's RC's
* OBOUND - Onset boundary descriptions
* AF - The analysis frame number
* Output:
* VOIBUF(2,0:AF) - Buffer of voicing decisions
* Internal:
* QS - Ratio of preemphasized to full-band energies
* RC1 - First reflection coefficient
* AR_B - Product of the causal forward and reverse pitch prediction gains
* AR_F - Product of the noncausal forward and rev. pitch prediction gains
* ZC - Zero crossing rate
* DITHER - Zero crossing threshold level
* MAXMIN - AMDF's 1 octave windowed maximum-to-minimum ratio
* MINPTR - Location of minimum AMDF value
* NVDC - Number of elements in each VDC vector
* NVDCL - Number of VDC vectors
* VDCL - SNR values corresponding to the set of VDC's
* VDC - 2-D voicing decision coefficient vector
* VALUE(9) - Voicing Parameters
* VOICE(2,3)- History of LDA results
* LBE - Ratio of low-band instantaneous to average energies
* FBE - Ratio of full-band instantaneous to average energies
* LBVE - Low band voiced energy
* LBUE - Low band unvoiced energy
* FBVE - Full band voiced energy
* FBUE - Full band unvoiced energy
* OFBUE - Previous full-band unvoiced energy
* OLBUE - Previous low-band unvoiced energy
* REF - Reference energy for initialization and DITHER threshold
* SNR - Estimate of signal-to-noise ratio
* SNR2 - Estimate of low-band signal-to-noise ratio
* SNRL - SNR level number
* OT - Onset transition present
* VSTATE - Decimal interpretation of binary voicing classifications
* FIRST - First call flag
*/
#include <stdio.h>
#include "vcomm.ch"
#include "contrl.ch"
#include "lpcdefs.h"
#include <math.h>
voicin( vwin, inbuf, lpbuf, half, minamd, maxamd, mintau, ivrc, obound, voibuf)
int vwin[2][AF], half, mintau;
float minamd, maxamd, ivrc[2];
float *inbuf;
float *lpbuf;
int *obound, voibuf[2][AF+1];
{
int zc, lbe, fbe;
int i, snrl;
static int vstate=0;
static float dither=20;
static float snr;
float snr2;
static float maxmin;
float qs, rc1, ar_b;
float ar_f;
static float voice[2][3];
float value[9];
short ot=0;
/* Declare and initialize filters: */
static int lbve, lbue, fbve, fbue, ofbue, olbue;
static int sfbue, slbue=0;
int ref= 3000;
static short first=1;
if (first) {
lbve = ref;
fbve = ref;
fbue = ref/16;
ofbue = ref/16;
lbue = ref/32;
olbue = ref/32;
snr = 64*(fbve/fbue);
first = 0;
vdcl[0] = 600;
vdcl[1] = 450;
vdcl[2] = 300;
vdcl[3] = 200;
vdcl[4] = 6*0;
for(i=0;i<3;i++) {
voice[1][i] = 0.0;
voice[0][i] = 0.0;
}
}
/* The VOICE array contains the result of the linear discriminant function
* (analog values). The VOIBUF array contains the hard-limited binary
* voicing decisions. The VOICE and VOIBUF arrays, according to FORTRAN
* memory allocation, are addressed as:
*
* (half-frame number, future-frame number)
*
* | Past | Present | Future1 | Future2 |
* | 1,0 | 2,0 | 1,1 | 2,1 | 1,2 | 2,2 | 1,3 | 2,3 | ---> time
*
* Update linear discriminant function history each frame: */
if (half == 1) {
voice[0][0]=voice[0][1];
voice[1][0]=voice[1][1];
voice[0][1]=voice[0][2];
voice[1][1]=voice[1][2];
maxmin = maxamd/mmax(minamd,1.);
}
/* Calculate voicing parameters twice per frame: */
vparms( vwin, inbuf, lpbuf, half, &dither, mintau, &zc, &lbe, &fbe, &qs, &rc1, &ar_b, &ar_f );
/* Estimate signal-to-noise ratio to select the appropriate VDC vector.
* The SNR is estimated as the running average of the ratio of the
* running average full-band voiced energy to the running average
* full-band unvoiced energy. SNR filter has gain of 63. */
snr = L_nint( 63*( snr + fbve/(float)(mmax(fbue,1)) )/64.);
snr2 = (snr*fbue)/mmax(lbue,1);
/* Quantize SNR to SNRL according to VDCL thresholds.*/
/*DO SNRL = 1, NVDCL-1 */
for (snrl=1;snrl<nvdcl;snrl++) {
if (snr2 > vdcl[snrl-1]) break;
}
/* (Note: SNRL = NVDCL Here) */
/* Linear discriminant voicing parameters: */
value[0] = maxmin;
value[1] = (float)(lbe)/mmax(lbve,1);
value[2] = zc;
value[3] = rc1;
value[4] = qs;
value[5] = ivrc[2];
value[6] = ar_b;
value[7] = ar_f;
value[8] = 0.0;
/* Evaluation of linear discriminant function: */
voice[half-1][2] = vdc[9][snrl-1];
for(i=1;i<10;i++) {
voice[half-1][2] += vdc[i-1][snrl-1]*value[i-1];
}
/* Classify as voiced if discriminant > 0, otherwise unvoiced
* Voicing decision for current half-frame: 1 = Voiced; 0 = Unvoiced */
if (voice[half-1][2] > 0.0)
voibuf[half-1][3]=1;
else
voibuf[half-1][3]=0;
/* Skip voicing decision smoothing in first half-frame: */
if (half != 1) {
/* Voicing decision smoothing rules (override of linear combination):
*
* Unvoiced half-frames: At least two in a row.
* --------------------
*
* Voiced half-frames: At least two in a row in one frame.
* ------------------- Otherwise at least three in a row.
* (Due to the way transition frames are encoded)
*
* In many cases, the discriminant function determines how to smooth.
* In the following chart, the decisions marked with a * may be overridden.
*
* Voicing override of transitions at onsets:
* If a V/UV or UV/V voicing decision transition occurs within one-half
* frame of an onset bounding a voicing window, then the transition is
* moved to occur at the onset.
*
* P 1F
* ----- -----
* 0 0 0 0
* 0 0 0* 1 (If there is an onset there)
* 0 0 1* 0* (Based on 2F and discriminant distance)
* 0 0 1 1
* 0 1* 0 0 (Always)
* 0 1* 0* 1 (Based on discriminant distance)
* 0* 1 1 0* (Based on past, 2F, and discriminant distance)
* 0 1* 1 1 (If there is an onset there)
* 1 0* 0 0 (If there is an onset there)
* 1 0 0 1
* 1 0* 1* 0 (Based on discriminant distance)
* 1 0* 1 1 (Always)
* 1 1 0 0
* 1 1 0* 1* (Based on 2F and discriminant distance)
* 1 1 1* 0 (If there is an onset there)
* 1 1 1 1
*
* Determine if there is an onset transition between P and 1F.
* OT (Onset Transition) is true if there is an onset between
* P and 1F but not after 1F.
*/
/*OT = (AND(OBOUND(1), 2) .NE. 0 .OR. OBOUND(2) .EQ. 1) .AND. AND(OBOUND(3), 1) .EQ. 0 */
ot = ((obound[1] & 2) != 0 || obound[2] == 1) && (obound[3] & 1) == 0;
/* Multi-way dispatch on voicing decision history: */
vstate = voibuf[0][1]*8 + voibuf[1][1]*4 + voibuf[0][2]*2 + voibuf[1][2];
/* GOTO (99,1,2,99,4,5,6,7,8,99,10,11,99,13,14,99) VSTATE+1 */
/*if(count==9) printf("vstate = %d\n",vstate);*/
switch(vstate+1) {
case 1:
break;
case 2:
if (ot && voibuf[0][3] == 1) voibuf[0][2] = 1;
break;
case 3:
if (voibuf[0][3] == 0 || voice[0][1] < -voice[1][1])
voibuf[0][2] = 0;
else
voibuf[1][2] = 1;
break;
case 4:
break;
case 5:
voibuf[1][1] = 0;
break;
case 6:
if (voice[1][0] < -voice[0][1])
voibuf[1][1] = 0;
else
voibuf[0][2] = 1;
break;
case 7:
/* VOIBUF(2,0) must be 0 */
if (voibuf[0][0] == 1 || voibuf[0][3] == 1 || voice[1][1] > voice[0][0])
voibuf[1][2] = 1;
else
voibuf[0][1] = 1;
break;
case 8:
if (ot) voibuf[1][1] = 0;
break;
case 9:
if (ot) voibuf[1][1] = 1;
break;
case 10:
break;
case 11:
if (voice[0][1] < -voice[1][0])
voibuf[0][2] = 0;
else
voibuf[1][1] = 1;
break;
case 12:
voibuf[1][1] = 1;
break;
case 13:
break;
case 14:
if ((voibuf[0][3] == 0) && (voice[1][1] < -voice[0][1]) )
voibuf[1][2] = 0;
else
voibuf[0][2] = 1;
break;
case 15:
if (ot && voibuf[0][3] == 0) voibuf[0][2] = 0;
break;
default:
break;
}
} /* (99)*/
/* Now update parameters:
* ----------------------
*
* During unvoiced half-frames, update the low band and full band unvoiced
* energy estimates (LBUE and FBUE) and also the zero crossing
* threshold (DITHER). (The input to the unvoiced energy filters is
* restricted to be less than 10dB above the previous inputs of the
* filters.)
* During voiced half-frames, update the low-pass (LBVE) and all-pass
* (FBVE) voiced energy estimates. */
if (voibuf[half-1][3] == 0) {
sfbue = L_nint(( 63*sfbue + 8*mmin(fbe,3*ofbue) )/64.);
fbue = sfbue/8;
ofbue = fbe;
slbue = L_nint(( 63*slbue + 8*mmin(lbe,3*olbue) )/64.);
lbue = slbue/8;
olbue = lbe;
}
else{
lbve = L_nint(( 63*lbve + lbe )/64.);
fbve = L_nint(( 63*fbve + fbe )/64.);
}
/* Set dither threshold to yield proper zero crossing rates in the
* presence of low frequency noise and low level signal input.
* NOTE: The divisor is a function of REF, the expected energies. */
dither = mmin(mmax( 64*sqrt((float)(lbue*lbve)) / ref,1.),20.);
/* Voicing decisions are returned in VOIBUF. */
}