1 /*
2
3 * Revision 1.2 1996/08/20 20:45:00 jaf
4 * Removed all static local variables that were SAVE'd in the Fortran
5 * code, and put them in struct lpc10_encoder_state that is passed as an
6 * argument.
7 *
8 * Removed init function, since all initialization is now done in
9 * init_lpc10_encoder_state().
10 *
11 * Revision 1.1 1996/08/19 22:30:14 jaf
12 * Initial revision
13 *
14
15 */
16
17 /* -- translated by f2c (version 19951025).
18 You must link the resulting object file with the libraries:
19 -lf2c -lm (in that order)
20 */
21
22 #include "f2c.h"
23
24 extern int voicin_(integer *vwin, real *inbuf, real *lpbuf, integer *buflim, integer *half, real *minamd, real *maxamd, integer *mintau, real *ivrc, integer *obound, integer *voibuf, integer *af, struct lpc10_encoder_state *st);
25
26 /* Common Block Declarations */
27
28 extern struct {
29 integer order, lframe;
30 logical corrp;
31 } contrl_;
32
33 #define contrl_1 contrl_
34
35 /****************************************************************************/
36
37 /* VOICIN Version 52 */
38
39 /*
40 * Revision 1.2 1996/08/20 20:45:00 jaf
41 * Removed all static local variables that were SAVE'd in the Fortran
42 * code, and put them in struct lpc10_encoder_state that is passed as an
43 * argument.
44 *
45 * Removed init function, since all initialization is now done in
46 * init_lpc10_encoder_state().
47 *
48 * Revision 1.1 1996/08/19 22:30:14 jaf
49 * Initial revision
50 * */
51 /* Revision 1.10 1996/03/29 17:59:14 jaf */
52 /* Avoided using VALUE(9), although it shouldn't affect the function of */
53 /* the code at all, because it was always multiplied by VDC(9,SNRL), */
54 /* which is 0 for all values of SNRL. Still, if VALUE(9) had an initial */
55 /* value of IEEE NaN, it might cause trouble (I don't know how IEEE */
56 /* defines Nan * 0. It should either be NaN or 0.) */
57
58 /* Revision 1.9 1996/03/29 17:54:46 jaf */
59 /* Added a few comments about the accesses made to argument array VOIBUF */
60 /* and the local saved array VOICE. */
61
62 /* Revision 1.8 1996/03/27 18:19:54 jaf */
63 /* Added an assignment to VSTATE that does not affect the function of the */
64 /* program at all. The only reason I put it in was so that the tracing */
65 /* statements at the end, when enabled, will print a consistent value for */
66 /* VSTATE when HALF .EQ. 1, rather than a garbage value that could change */
67 /* from one call to the next. */
68
69 /* Revision 1.7 1996/03/26 20:00:06 jaf */
70 /* Removed the inclusion of the file "vcomm.fh", and put its contents */
71 /* into this file. It was included nowhere else but here. */
72
73 /* Revision 1.6 1996/03/26 19:38:09 jaf */
74 /* Commented out trace statements. */
75
76 /* Revision 1.5 1996/03/19 20:43:45 jaf */
77 /* Added comments about which indices of OBOUND and VOIBUF can be */
78 /* accessed, and whether they are read or written. VOIBUF is fairly */
79 /* messy. */
80
81 /* Revision 1.4 1996/03/19 15:00:58 jaf */
82 /* Moved the DATA statements for the *VDC* variables later, as it is */
83 /* apparently illegal to have DATA statements before local variable */
84 /* declarations. */
85
86 /* Revision 1.3 1996/03/19 00:10:49 jaf */
87 /* Heavily commented the local variables that are saved from one */
88 /* invocation to the next, and how the local variable FIRST is used to */
89 /* avoid the need to assign most of them initial values with DATA */
90 /* statements. */
91
92 /* A few should be initialized, but aren't. I've guessed initial values */
93 /* for two of these, SFBUE and SLBUE, and I've convinced myself that for */
94 /* VOICE, the effects of uninitialized values will die out after 2 or 3 */
95 /* frame times. It would still be good to choose initial values for */
96 /* these, but I don't know what reasonable values would be (0 comes to */
97 /* mind). */
98
99 /* Revision 1.2 1996/03/13 16:09:28 jaf */
100 /* Comments added explaining which of the local variables of this */
101 /* subroutine need to be saved from one invocation to the next, and which */
102 /* do not. */
103
104 /* WARNING! Some of them that should are never given initial values in */
105 /* this code. Hopefully, Fortran 77 defines initial values for them, but */
106 /* even so, giving them explicit initial values is preferable. */
107
108 /* WARNING! VALUE(9) is used, but never assigned a value. It should */
109 /* probably be eliminated from the code. */
110
111 /* Revision 1.1 1996/02/07 14:50:28 jaf */
112 /* Initial revision */
113
114
115 /****************************************************************************/
116
117 /* Voicing Detection (VOICIN) makes voicing decisions for each half */
118 /* frame of input speech. Tentative voicing decisions are made two frames*/
119 /* in the future (2F) for each half frame. These decisions are carried */
120 /* through one frame in the future (1F) to the present (P) frame where */
121 /* they are examined and smoothed, resulting in the final voicing */
122 /* decisions for each half frame. */
123 /* The voicing parameter (signal measurement) column vector (VALUE) */
124 /* is based on a rectangular window of speech samples determined by the */
125 /* window placement algorithm. The voicing parameter vector contains the*/
126 /* AMDF windowed maximum-to-minimum ratio, the zero crossing rate, energy*/
127 /* measures, reflection coefficients, and prediction gains. The voicing */
128 /* window is placed to avoid contamination of the voicing parameter vector*/
129 /* with speech onsets. */
130 /* The input signal is then classified as unvoiced (including */
131 /* silence) or voiced. This decision is made by a linear discriminant */
132 /* function consisting of a dot product of the voicing decision */
133 /* coefficient (VDC) row vector with the measurement column vector */
134 /* (VALUE). The VDC vector is 2-dimensional, each row vector is optimized*/
135 /* for a particular signal-to-noise ratio (SNR). So, before the dot */
136 /* product is performed, the SNR is estimated to select the appropriate */
137 /* VDC vector. */
138 /* The smoothing algorithm is a modified median smoother. The */
139 /* voicing discriminant function is used by the smoother to determine how*/
140 /* strongly voiced or unvoiced a signal is. The smoothing is further */
141 /* modified if a speech onset and a voicing decision transition occur */
142 /* within one half frame. In this case, the voicing decision transition */
143 /* is extended to the speech onset. For transmission purposes, there are*/
144 /* constraints on the duration and transition of voicing decisions. The */
145 /* smoother takes these constraints into account. */
146 /* Finally, the energy estimates are updated along with the dither */
147 /* threshold used to calculate the zero crossing rate (ZC). */
148
149 /* Inputs: */
150 /* VWIN - Voicing window limits */
151 /* The indices read of arrays VWIN, INBUF, LPBUF, and BUFLIM */
152 /* are the same as those read by subroutine VPARMS. */
153 /* INBUF - Input speech buffer */
154 /* LPBUF - Low-pass filtered speech buffer */
155 /* BUFLIM - INBUF and LPBUF limits */
156 /* HALF - Present analysis half frame number */
157 /* MINAMD - Minimum value of the AMDF */
158 /* MAXAMD - Maximum value of the AMDF */
159 /* MINTAU - Pointer to the lag of the minimum AMDF value */
160 /* IVRC(2) - Inverse filter's RC's */
161 /* Only index 2 of array IVRC read under normal operation. */
162 /* (Index 1 is also read when debugging is turned on.) */
163 /* OBOUND - Onset boundary descriptions */
164 /* Indices 1 through 3 read if (HALF .NE. 1), otherwise untouched.
165 */
166 /* AF - The analysis frame number */
167 /* Output: */
168 /* VOIBUF(2,0:AF) - Buffer of voicing decisions */
169 /* Index (HALF,3) written. */
170 /* If (HALF .EQ. 1), skip down to "Read (HALF,3)" below. */
171 /* Indices (1,2), (2,1), (1,2), and (2,2) read. */
172 /* One of the following is then done: */
173 /* read (1,3) and possibly write (1,2) */
174 /* read (1,3) and write (1,2) or (2,2) */
175 /* write (2,1) */
176 /* write (2,1) or (1,2) */
177 /* read (1,0) and (1,3) and then write (2,2) or (1,1) */
178 /* no reads or writes on VOIBUF */
179 /* Finally, read (HALF,3) */
180 /* Internal: */
181 /* QS - Ratio of preemphasized to full-band energies */
182 /* RC1 - First reflection coefficient */
183 /* AR_B - Product of the causal forward and reverse pitch prediction gain
184 s*/
185 /* AR_F - Product of the noncausal forward and rev. pitch prediction gain
186 s*/
187 /* ZC - Zero crossing rate */
188 /* DITHER - Zero crossing threshold level */
189 /* MAXMIN - AMDF's 1 octave windowed maximum-to-minimum ratio */
190 /* MINPTR - Location of minimum AMDF value */
191 /* NVDC - Number of elements in each VDC vector */
192 /* NVDCL - Number of VDC vectors */
193 /* VDCL - SNR values corresponding to the set of VDC's */
194 /* VDC - 2-D voicing decision coefficient vector */
195 /* VALUE(9) - Voicing Parameters */
196 /* VOICE(2,3)- History of LDA results */
197 /* On every call when (HALF .EQ. 1), VOICE(*,I+1) is */
198 /* shifted back to VOICE(*,I), for I=1,2. */
199 /* VOICE(HALF,3) is written on every call. */
200 /* Depending on several conditions, one or more of */
201 /* (1,1), (1,2), (2,1), and (2,2) might then be read. */
202 /* LBE - Ratio of low-band instantaneous to average energies */
203 /* FBE - Ratio of full-band instantaneous to average energies */
204 /* LBVE - Low band voiced energy */
205 /* LBUE - Low band unvoiced energy */
206 /* FBVE - Full band voiced energy */
207 /* FBUE - Full band unvoiced energy */
208 /* OFBUE - Previous full-band unvoiced energy */
209 /* OLBUE - Previous low-band unvoiced energy */
210 /* REF - Reference energy for initialization and DITHER threshold */
211 /* SNR - Estimate of signal-to-noise ratio */
212 /* SNR2 - Estimate of low-band signal-to-noise ratio */
213 /* SNRL - SNR level number */
214 /* OT - Onset transition present */
215 /* VSTATE - Decimal interpretation of binary voicing classifications */
216 /* FIRST - First call flag */
217
218 /* This subroutine maintains local state from one call to the next. If */
219 /* you want to switch to using a new audio stream for this filter, or */
220 /* reinitialize its state for any other reason, call the ENTRY */
221 /* INITVOICIN. */
222
voicin_(integer * vwin,real * inbuf,real * lpbuf,integer * buflim,integer * half,real * minamd,real * maxamd,integer * mintau,real * ivrc,integer * obound,integer * voibuf,integer * af,struct lpc10_encoder_state * st)223 /* Subroutine */ int voicin_(integer *vwin, real *inbuf, real *
224 lpbuf, integer *buflim, integer *half, real *minamd, real *maxamd,
225 integer *mintau, real *ivrc, integer *obound, integer *voibuf,
226 integer *af, struct lpc10_encoder_state *st)
227 {
228 /* Initialized data */
229
230 real *dither;
231 static real vdc[100] /* was [10][10] */ = { 0.f,1714.f,-110.f,
232 334.f,-4096.f,-654.f,3752.f,3769.f,0.f,1181.f,0.f,874.f,-97.f,
233 300.f,-4096.f,-1021.f,2451.f,2527.f,0.f,-500.f,0.f,510.f,-70.f,
234 250.f,-4096.f,-1270.f,2194.f,2491.f,0.f,-1500.f,0.f,500.f,-10.f,
235 200.f,-4096.f,-1300.f,2e3f,2e3f,0.f,-2e3f,0.f,500.f,0.f,0.f,
236 -4096.f,-1300.f,2e3f,2e3f,0.f,-2500.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,
237 0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,
238 0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,
239 0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f };
240 static integer nvdcl = 5;
241 static real vdcl[10] = { 600.f,450.f,300.f,200.f,0.f,0.f,0.f,0.f,0.f,0.f }
242 ;
243
244 /* System generated locals */
245 integer inbuf_offset = 0, lpbuf_offset = 0, i__1, i__2;
246 real r__1, r__2;
247
248 /* Builtin functions */
249 integer i_nint(real *);
250 double sqrt(doublereal);
251
252 /* Local variables */
253 real ar_b__, ar_f__;
254 integer *lbve, *lbue, *fbve, *fbue;
255 integer snrl, i__;
256 integer *ofbue, *sfbue;
257 real *voice;
258 integer *olbue, *slbue;
259 real value[9];
260 integer zc;
261 logical ot;
262 real qs;
263 real *maxmin;
264 integer vstate;
265 real rc1;
266 extern /* Subroutine */ int vparms_(integer *, real *, real *, integer *,
267 integer *, real *, integer *, integer *, integer *, integer *,
268 real *, real *, real *, real *);
269 integer fbe, lbe;
270 real *snr;
271 real snr2;
272
273 (void)af;
274 /* Global Variables: */
275 /* Arguments */
276
277 /* LPC Processing control variables: */
278
279 /* *** Read-only: initialized in setup */
280
281 /* Files for Speech, Parameter, and Bitstream Input & Output, */
282 /* and message and debug outputs. */
283
284 /* Here are the only files which use these variables: */
285
286 /* lpcsim.f setup.f trans.f error.f vqsetup.f */
287
288 /* Many files which use fdebug are not listed, since it is only used in */
289 /* those other files conditionally, to print trace statements. */
290 /* integer fsi, fso, fpi, fpo, fbi, fbo, pbin, fmsg, fdebug */
291 /* LPC order, Frame size, Quantization rate, Bits per frame, */
292 /* Error correction */
293 /* Subroutine SETUP is the only place where order is assigned a value, */
294 /* and that value is 10. It could increase efficiency 1% or so to */
295 /* declare order as a constant (i.e., a Fortran PARAMETER) instead of as
296 */
297 /* a variable in a COMMON block, since it is used in many places in the */
298 /* core of the coding and decoding routines. Actually, I take that back.
299 */
300 /* At least when compiling with f2c, the upper bound of DO loops is */
301 /* stored in a local variable before the DO loop begins, and then that is
302 */
303 /* compared against on each iteration. */
304 /* Similarly for lframe, which is given a value of MAXFRM in SETUP. */
305 /* Similarly for quant, which is given a value of 2400 in SETUP. quant */
306 /* is used in only a few places, and never in the core coding and */
307 /* decoding routines, so it could be eliminated entirely. */
308 /* nbits is similar to quant, and is given a value of 54 in SETUP. */
309 /* corrp is given a value of .TRUE. in SETUP, and is only used in the */
310 /* subroutines ENCODE and DECODE. It doesn't affect the speed of the */
311 /* coder significantly whether it is .TRUE. or .FALSE., or whether it is
312 */
313 /* a constant or a variable, since it is only examined once per frame. */
314 /* Leaving it as a variable that is set to .TRUE. seems like a good */
315 /* idea, since it does enable some error-correction capability for */
316 /* unvoiced frames, with no change in the coding rate, and no noticeable
317 */
318 /* quality difference in the decoded speech. */
319 /* integer quant, nbits */
320 /* *** Read/write: variables for debugging, not needed for LPC algorithm
321 */
322
323 /* Current frame, Unstable frames, Output clip count, Max onset buffer,
324 */
325 /* Debug listing detail level, Line count on listing page */
326
327 /* nframe is not needed for an embedded LPC10 at all. */
328 /* nunsfm is initialized to 0 in SETUP, and incremented in subroutine */
329 /* ERROR, which is only called from RCCHK. When LPC10 is embedded into */
330 /* an application, I would recommend removing the call to ERROR in RCCHK,
331 */
332 /* and remove ERROR and nunsfm completely. */
333 /* iclip is initialized to 0 in SETUP, and incremented in entry SWRITE in
334 */
335 /* sread.f. When LPC10 is embedded into an application, one might want */
336 /* to cause it to be incremented in a routine that takes the output of */
337 /* SYNTHS and sends it to an audio device. It could be optionally */
338 /* displayed, for those that might want to know what it is. */
339 /* maxosp is never initialized to 0 in SETUP, although it probably should
340 */
341 /* be, and it is updated in subroutine ANALYS. I doubt that its value */
342 /* would be of much interest to an application in which LPC10 is */
343 /* embedded. */
344 /* listl and lincnt are not needed for an embedded LPC10 at all. */
345 /* integer nframe, nunsfm, iclip, maxosp, listl, lincnt */
346 /* common /contrl/ fsi, fso, fpi, fpo, fbi, fbo, pbin, fmsg, fdebug */
347 /* common /contrl/ quant, nbits */
348 /* common /contrl/ nframe, nunsfm, iclip, maxosp, listl, lincnt */
349 /* Parameters/constants */
350 /* Voicing coefficient and Linear Discriminant Analysis variables:
351 */
352 /* Max number of VDC's and VDC levels */
353 /* The following are not Fortran PARAMETER's, but they are */
354 /* initialized with DATA statements, and never modified. */
355 /* Actual number of VDC's and levels */
356 /* Local variables that need not be saved */
357 /* Note: */
358
359 /* VALUE(1) through VALUE(8) are assigned values, but VALUE(9) */
360 /* never is. Yet VALUE(9) is read in the loop that begins "DO I =
361 */
362 /* 1, 9" below. I believe that this doesn't cause any problems in
363 */
364 /* this subroutine, because all VDC(9,*) array elements are 0, and
365 */
366 /* this is what is multiplied by VALUE(9) in all cases. Still, it
367 */
368 /* would save a multiplication to change the loop to "DO I = 1, 8".
369 */
370 /* Local state */
371 /* WARNING! */
372
373 /* VOICE, SFBUE, and SLBUE should be saved from one invocation to */
374 /* the next, but they are never given an initial value. */
375
376 /* Does Fortran 77 specify some default initial value, like 0, or */
377 /* is it undefined? If it is undefined, then this code should be */
378 /* corrected to specify an initial value. */
379
380 /* For VOICE, note that it is "shifted" in the statement that */
381 /* begins "IF (HALF .EQ. 1) THEN" below. Also, uninitialized */
382 /* values in the VOICE array can only affect entries in the VOIBUF
383 */
384 /* array that are for the same frame, or for an older frame. Thus
385 */
386 /* the effects of uninitialized values in VOICE cannot linger on */
387 /* for more than 2 or 3 frame times. */
388
389 /* For SFBUE and SLBUE, the effects of uninitialized values can */
390 /* linger on for many frame times, because their previous values */
391 /* are exponentially decayed. Thus it is more important to choose
392 */
393 /* initial values for these variables. I would guess that a */
394 /* reasonable initial value for SFBUE is REF/16, the same as used */
395 /* for FBUE and OFBUE. Similarly, SLBUE can be initialized to */
396 /* REF/32, the same as for LBUE and OLBUE. */
397
398 /* These guessed initial values should be validated by re-running */
399 /* the modified program on some audio samples. */
400
401 /* Declare and initialize filters: */
402
403 dither = (&st->dither);
404 snr = (&st->snr);
405 maxmin = (&st->maxmin);
406 voice = (&st->voice[0]);
407 lbve = (&st->lbve);
408 lbue = (&st->lbue);
409 fbve = (&st->fbve);
410 fbue = (&st->fbue);
411 ofbue = (&st->ofbue);
412 olbue = (&st->olbue);
413 sfbue = (&st->sfbue);
414 slbue = (&st->slbue);
415
416 /* Parameter adjustments */
417 if (vwin) {
418 --vwin;
419 }
420 if (buflim) {
421 --buflim;
422 }
423 if (inbuf) {
424 inbuf_offset = buflim[1];
425 inbuf -= inbuf_offset;
426 }
427 if (lpbuf) {
428 lpbuf_offset = buflim[3];
429 lpbuf -= lpbuf_offset;
430 }
431 if (ivrc) {
432 --ivrc;
433 }
434 if (obound) {
435 --obound;
436 }
437 if (voibuf) {
438 --voibuf;
439 }
440
441 /* Function Body */
442
443 /* The following variables are saved from one invocation to the */
444 /* next, but are not initialized with DATA statements. This is */
445 /* acceptable, because FIRST is initialized ot .TRUE., and the */
446 /* first time that this subroutine is then called, they are all */
447 /* given initial values. */
448
449 /* SNR */
450 /* LBVE, LBUE, FBVE, FBUE, OFBUE, OLBUE */
451
452 /* MAXMIN is initialized on the first call, assuming that HALF */
453 /* .EQ. 1 on first call. This is how ANALYS calls this subroutine.
454 */
455
456 /* Voicing Decision Parameter vector (* denotes zero coefficient): */
457
458 /* * MAXMIN */
459 /* LBE/LBVE */
460 /* ZC */
461 /* RC1 */
462 /* QS */
463 /* IVRC2 */
464 /* aR_B */
465 /* aR_F */
466 /* * LOG(LBE/LBVE) */
467 /* Define 2-D voicing decision coefficient vector according to the voicin
468 g*/
469 /* parameter order above. Each row (VDC vector) is optimized for a speci
470 fic*/
471 /* SNR. The last element of the vector is the constant. */
472 /* E ZC RC1 Qs IVRC2 aRb aRf c */
473
474 /* The VOICE array contains the result of the linear discriminant functio
475 n*/
476 /* (analog values). The VOIBUF array contains the hard-limited binary
477 */
478 /* voicing decisions. The VOICE and VOIBUF arrays, according to FORTRAN
479 */
480 /* memory allocation, are addressed as: */
481
482 /* (half-frame number, future-frame number) */
483
484 /* | Past | Present | Future1 | Future2 | */
485 /* | 1,0 | 2,0 | 1,1 | 2,1 | 1,2 | 2,2 | 1,3 | 2,3 | ---> time */
486
487 /* Update linear discriminant function history each frame: */
488 if (*half == 1) {
489 voice[0] = voice[2];
490 voice[1] = voice[3];
491 voice[2] = voice[4];
492 voice[3] = voice[5];
493 *maxmin = *maxamd / max(*minamd,1.f);
494 }
495 /* Calculate voicing parameters twice per frame: */
496 vparms_(&vwin[1], &inbuf[inbuf_offset], &lpbuf[lpbuf_offset], &buflim[1],
497 half, dither, mintau, &zc, &lbe, &fbe, &qs, &rc1, &ar_b__, &
498 ar_f__);
499 /* Estimate signal-to-noise ratio to select the appropriate VDC vector.
500 */
501 /* The SNR is estimated as the running average of the ratio of the */
502 /* running average full-band voiced energy to the running average */
503 /* full-band unvoiced energy. SNR filter has gain of 63. */
504 r__1 = (*snr + *fbve / (real) max(*fbue,1)) * 63 / 64.f;
505 *snr = (real) i_nint(&r__1);
506 snr2 = *snr * *fbue / max(*lbue,1);
507 /* Quantize SNR to SNRL according to VDCL thresholds. */
508 snrl = 1;
509 i__1 = nvdcl - 1;
510 for (snrl = 1; snrl <= i__1; ++snrl) {
511 if (snr2 > vdcl[snrl - 1]) {
512 goto L69;
513 }
514 }
515 /* (Note: SNRL = NVDCL here) */
516 L69:
517 /* Linear discriminant voicing parameters: */
518 value[0] = *maxmin;
519 value[1] = (real) lbe / max(*lbve,1);
520 value[2] = (real) zc;
521 value[3] = rc1;
522 value[4] = qs;
523 value[5] = ivrc[2];
524 value[6] = ar_b__;
525 value[7] = ar_f__;
526 /* Evaluation of linear discriminant function: */
527 voice[*half + 3] = vdc[snrl * 10 - 1];
528 for (i__ = 1; i__ <= 8; ++i__) {
529 voice[*half + 3] += vdc[i__ + snrl * 10 - 11] * value[i__ - 1];
530 }
531 /* Classify as voiced if discriminant > 0, otherwise unvoiced */
532 /* Voicing decision for current half-frame: 1 = Voiced; 0 = Unvoiced */
533 if (voice[*half + 3] > 0.f) {
534 voibuf[*half + 6] = 1;
535 } else {
536 voibuf[*half + 6] = 0;
537 }
538 /* Skip voicing decision smoothing in first half-frame: */
539 /* Give a value to VSTATE, so that trace statements below will print
540 */
541 /* a consistent value from one call to the next when HALF .EQ. 1. */
542 /* The value of VSTATE is not used for any other purpose when this is
543 */
544 /* true. */
545 vstate = -1;
546 if (*half == 1) {
547 goto L99;
548 }
549 /* Voicing decision smoothing rules (override of linear combination): */
550
551 /* Unvoiced half-frames: At least two in a row. */
552 /* -------------------- */
553
554 /* Voiced half-frames: At least two in a row in one frame. */
555 /* ------------------- Otherwise at least three in a row. */
556 /* (Due to the way transition frames are encoded) */
557
558 /* In many cases, the discriminant function determines how to smooth. */
559 /* In the following chart, the decisions marked with a * may be overridden
560 .*/
561
562 /* Voicing override of transitions at onsets: */
563 /* If a V/UV or UV/V voicing decision transition occurs within one-half
564 */
565 /* frame of an onset bounding a voicing window, then the transition is */
566 /* moved to occur at the onset. */
567
568 /* P 1F */
569 /* ----- ----- */
570 /* 0 0 0 0 */
571 /* 0 0 0* 1 (If there is an onset there) */
572 /* 0 0 1* 0* (Based on 2F and discriminant distance) */
573 /* 0 0 1 1 */
574 /* 0 1* 0 0 (Always) */
575 /* 0 1* 0* 1 (Based on discriminant distance) */
576 /* 0* 1 1 0* (Based on past, 2F, and discriminant distance) */
577 /* 0 1* 1 1 (If there is an onset there) */
578 /* 1 0* 0 0 (If there is an onset there) */
579 /* 1 0 0 1 */
580 /* 1 0* 1* 0 (Based on discriminant distance) */
581 /* 1 0* 1 1 (Always) */
582 /* 1 1 0 0 */
583 /* 1 1 0* 1* (Based on 2F and discriminant distance) */
584 /* 1 1 1* 0 (If there is an onset there) */
585 /* 1 1 1 1 */
586
587 /* Determine if there is an onset transition between P and 1F. */
588 /* OT (Onset Transition) is true if there is an onset between */
589 /* P and 1F but not after 1F. */
590 ot = ((obound[1] & 2) != 0 || obound[2] == 1) && (obound[3] & 1) == 0;
591 /* Multi-way dispatch on voicing decision history: */
592 vstate = (voibuf[3] << 3) + (voibuf[4] << 2) + (voibuf[5] << 1) + voibuf[
593 6];
594 switch (vstate + 1) {
595 case 1: goto L99;
596 case 2: goto L1;
597 case 3: goto L2;
598 case 4: goto L99;
599 case 5: goto L4;
600 case 6: goto L5;
601 case 7: goto L6;
602 case 8: goto L7;
603 case 9: goto L8;
604 case 10: goto L99;
605 case 11: goto L10;
606 case 12: goto L11;
607 case 13: goto L99;
608 case 14: goto L13;
609 case 15: goto L14;
610 case 16: goto L99;
611 }
612 L1:
613 if (ot && voibuf[7] == 1) {
614 voibuf[5] = 1;
615 }
616 goto L99;
617 L2:
618 if (voibuf[7] == 0 || voice[2] < -voice[3]) {
619 voibuf[5] = 0;
620 } else {
621 voibuf[6] = 1;
622 }
623 goto L99;
624 L4:
625 voibuf[4] = 0;
626 goto L99;
627 L5:
628 if (voice[1] < -voice[2]) {
629 voibuf[4] = 0;
630 } else {
631 voibuf[5] = 1;
632 }
633 goto L99;
634 /* VOIBUF(2,0) must be 0 */
635 L6:
636 if (voibuf[1] == 1 || voibuf[7] == 1 || voice[3] > voice[0]) {
637 voibuf[6] = 1;
638 } else {
639 voibuf[3] = 1;
640 }
641 goto L99;
642 L7:
643 if (ot) {
644 voibuf[4] = 0;
645 }
646 goto L99;
647 L8:
648 if (ot) {
649 voibuf[4] = 1;
650 }
651 goto L99;
652 L10:
653 if (voice[2] < -voice[1]) {
654 voibuf[5] = 0;
655 } else {
656 voibuf[4] = 1;
657 }
658 goto L99;
659 L11:
660 voibuf[4] = 1;
661 goto L99;
662 L13:
663 if (voibuf[7] == 0 && voice[3] < -voice[2]) {
664 voibuf[6] = 0;
665 } else {
666 voibuf[5] = 1;
667 }
668 goto L99;
669 L14:
670 if (ot && voibuf[7] == 0) {
671 voibuf[5] = 0;
672 }
673 /* GOTO 99 */
674 L99:
675 /* Now update parameters: */
676 /* ---------------------- */
677
678 /* During unvoiced half-frames, update the low band and full band unvoice
679 d*/
680 /* energy estimates (LBUE and FBUE) and also the zero crossing */
681 /* threshold (DITHER). (The input to the unvoiced energy filters is */
682 /* restricted to be less than 10dB above the previous inputs of the */
683 /* filters.) */
684 /* During voiced half-frames, update the low-pass (LBVE) and all-pass */
685 /* (FBVE) voiced energy estimates. */
686 if (voibuf[*half + 6] == 0) {
687 /* Computing MIN */
688 i__1 = fbe, i__2 = *ofbue * 3;
689 r__1 = (*sfbue * 63 + (min(i__1,i__2) << 3)) / 64.f;
690 *sfbue = i_nint(&r__1);
691 *fbue = *sfbue / 8;
692 *ofbue = fbe;
693 /* Computing MIN */
694 i__1 = lbe, i__2 = *olbue * 3;
695 r__1 = (*slbue * 63 + (min(i__1,i__2) << 3)) / 64.f;
696 *slbue = i_nint(&r__1);
697 *lbue = *slbue / 8;
698 *olbue = lbe;
699 } else {
700 r__1 = (*lbve * 63 + lbe) / 64.f;
701 *lbve = i_nint(&r__1);
702 r__1 = (*fbve * 63 + fbe) / 64.f;
703 *fbve = i_nint(&r__1);
704 }
705 /* Set dither threshold to yield proper zero crossing rates in the */
706 /* presence of low frequency noise and low level signal input. */
707 /* NOTE: The divisor is a function of REF, the expected energies. */
708 /* Computing MIN */
709 /* Computing MAX */
710 r__2 = sqrt((real) (*lbue * *lbve)) * 64 / 3000;
711 r__1 = max(r__2,1.f);
712 *dither = min(r__1,20.f);
713 /* Voicing decisions are returned in VOIBUF. */
714 return 0;
715 } /* voicin_ */
716