1 /*
2
3 $Log$
4 Revision 1.2 2006/08/01 13:06:51 rjongbloed
5 Added a raft of unvalidated audio codecs from OpenH323 tree
6
7 Revision 1.1.2.1 2006/07/22 14:03:19 rjongbloed
8 Added more plug ins
9
10 Revision 1.1.2.1 2006/05/08 13:49:58 rjongbloed
11 Imported all the audio codec plug ins from OpenH323
12
13 Revision 1.1 2004/05/04 11:16:43 csoutheren
14 Initial version
15
16 Revision 1.2 2002/02/15 03:57:55 yurik
17 Warnings removed during compilation, patch courtesy of Jehan Bing, jehan@bravobrava.com
18
19 Revision 1.1 2000/06/05 04:45:12 robertj
20 Added LPC-10 2400bps codec
21
22 * Revision 1.2 1996/08/20 20:45:00 jaf
23 * Removed all static local variables that were SAVE'd in the Fortran
24 * code, and put them in struct lpc10_encoder_state that is passed as an
25 * argument.
26 *
27 * Removed init function, since all initialization is now done in
28 * init_lpc10_encoder_state().
29 *
30 * Revision 1.1 1996/08/19 22:30:14 jaf
31 * Initial revision
32 *
33
34 */
35
36 #ifdef P_R_O_T_O_T_Y_P_E_S
37 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);
38 /* comlen contrl_ 12 */
39 /*:ref: vparms_ 14 14 4 6 6 4 4 6 4 4 4 4 6 6 6 6 */
40 #endif
41
42 /* -- translated by f2c (version 19951025).
43 You must link the resulting object file with the libraries:
44 -lf2c -lm (in that order)
45 */
46
47 #include "f2c.h"
48
49 /* Common Block Declarations */
50
51 extern struct {
52 integer order, lframe;
53 logical corrp;
54 } contrl_;
55
56 #define contrl_1 contrl_
57
58 /****************************************************************************/
59
60 /* VOICIN Version 52 */
61
62 /* $Log$
63 /* Revision 1.2 2006/08/01 13:06:51 rjongbloed
64 /* Added a raft of unvalidated audio codecs from OpenH323 tree
65 /*
66 /* Revision 1.1.2.1 2006/07/22 14:03:19 rjongbloed
67 /* Added more plug ins
68 /*
69 /* Revision 1.1.2.1 2006/05/08 13:49:58 rjongbloed
70 /* Imported all the audio codec plug ins from OpenH323
71 /*
72 /* Revision 1.1 2004/05/04 11:16:43 csoutheren
73 /* Initial version
74 /*
75 /* Revision 1.2 2002/02/15 03:57:55 yurik
76 /* Warnings removed during compilation, patch courtesy of Jehan Bing, jehan@bravobrava.com
77 /*
78 /* Revision 1.1 2000/06/05 04:45:12 robertj
79 /* Added LPC-10 2400bps codec
80 /*
81 * Revision 1.2 1996/08/20 20:45:00 jaf
82 * Removed all static local variables that were SAVE'd in the Fortran
83 * code, and put them in struct lpc10_encoder_state that is passed as an
84 * argument.
85 *
86 * Removed init function, since all initialization is now done in
87 * init_lpc10_encoder_state().
88 *
89 * Revision 1.1 1996/08/19 22:30:14 jaf
90 * Initial revision
91 * */
92 /* Revision 1.10 1996/03/29 17:59:14 jaf */
93 /* Avoided using VALUE(9), although it shouldn't affect the function of */
94 /* the code at all, because it was always multiplied by VDC(9,SNRL), */
95 /* which is 0 for all values of SNRL. Still, if VALUE(9) had an initial */
96 /* value of IEEE NaN, it might cause trouble (I don't know how IEEE */
97 /* defines Nan * 0. It should either be NaN or 0.) */
98
99 /* Revision 1.9 1996/03/29 17:54:46 jaf */
100 /* Added a few comments about the accesses made to argument array VOIBUF */
101 /* and the local saved array VOICE. */
102
103 /* Revision 1.8 1996/03/27 18:19:54 jaf */
104 /* Added an assignment to VSTATE that does not affect the function of the */
105 /* program at all. The only reason I put it in was so that the tracing */
106 /* statements at the end, when enabled, will print a consistent value for */
107 /* VSTATE when HALF .EQ. 1, rather than a garbage value that could change */
108 /* from one call to the next. */
109
110 /* Revision 1.7 1996/03/26 20:00:06 jaf */
111 /* Removed the inclusion of the file "vcomm.fh", and put its contents */
112 /* into this file. It was included nowhere else but here. */
113
114 /* Revision 1.6 1996/03/26 19:38:09 jaf */
115 /* Commented out trace statements. */
116
117 /* Revision 1.5 1996/03/19 20:43:45 jaf */
118 /* Added comments about which indices of OBOUND and VOIBUF can be */
119 /* accessed, and whether they are read or written. VOIBUF is fairly */
120 /* messy. */
121
122 /* Revision 1.4 1996/03/19 15:00:58 jaf */
123 /* Moved the DATA statements for the *VDC* variables later, as it is */
124 /* apparently illegal to have DATA statements before local variable */
125 /* declarations. */
126
127 /* Revision 1.3 1996/03/19 00:10:49 jaf */
128 /* Heavily commented the local variables that are saved from one */
129 /* invocation to the next, and how the local variable FIRST is used to */
130 /* avoid the need to assign most of them initial values with DATA */
131 /* statements. */
132
133 /* A few should be initialized, but aren't. I've guessed initial values */
134 /* for two of these, SFBUE and SLBUE, and I've convinced myself that for */
135 /* VOICE, the effects of uninitialized values will die out after 2 or 3 */
136 /* frame times. It would still be good to choose initial values for */
137 /* these, but I don't know what reasonable values would be (0 comes to */
138 /* mind). */
139
140 /* Revision 1.2 1996/03/13 16:09:28 jaf */
141 /* Comments added explaining which of the local variables of this */
142 /* subroutine need to be saved from one invocation to the next, and which */
143 /* do not. */
144
145 /* WARNING! Some of them that should are never given initial values in */
146 /* this code. Hopefully, Fortran 77 defines initial values for them, but */
147 /* even so, giving them explicit initial values is preferable. */
148
149 /* WARNING! VALUE(9) is used, but never assigned a value. It should */
150 /* probably be eliminated from the code. */
151
152 /* Revision 1.1 1996/02/07 14:50:28 jaf */
153 /* Initial revision */
154
155
156 /****************************************************************************/
157
158 /* Voicing Detection (VOICIN) makes voicing decisions for each half */
159 /* frame of input speech. Tentative voicing decisions are made two frames*/
160 /* in the future (2F) for each half frame. These decisions are carried */
161 /* through one frame in the future (1F) to the present (P) frame where */
162 /* they are examined and smoothed, resulting in the final voicing */
163 /* decisions for each half frame. */
164 /* The voicing parameter (signal measurement) column vector (VALUE) */
165 /* is based on a rectangular window of speech samples determined by the */
166 /* window placement algorithm. The voicing parameter vector contains the*/
167 /* AMDF windowed maximum-to-minimum ratio, the zero crossing rate, energy*/
168 /* measures, reflection coefficients, and prediction gains. The voicing */
169 /* window is placed to avoid contamination of the voicing parameter vector*/
170 /* with speech onsets. */
171 /* The input signal is then classified as unvoiced (including */
172 /* silence) or voiced. This decision is made by a linear discriminant */
173 /* function consisting of a dot product of the voicing decision */
174 /* coefficient (VDC) row vector with the measurement column vector */
175 /* (VALUE). The VDC vector is 2-dimensional, each row vector is optimized*/
176 /* for a particular signal-to-noise ratio (SNR). So, before the dot */
177 /* product is performed, the SNR is estimated to select the appropriate */
178 /* VDC vector. */
179 /* The smoothing algorithm is a modified median smoother. The */
180 /* voicing discriminant function is used by the smoother to determine how*/
181 /* strongly voiced or unvoiced a signal is. The smoothing is further */
182 /* modified if a speech onset and a voicing decision transition occur */
183 /* within one half frame. In this case, the voicing decision transition */
184 /* is extended to the speech onset. For transmission purposes, there are*/
185 /* constraints on the duration and transition of voicing decisions. The */
186 /* smoother takes these constraints into account. */
187 /* Finally, the energy estimates are updated along with the dither */
188 /* threshold used to calculate the zero crossing rate (ZC). */
189
190 /* Inputs: */
191 /* VWIN - Voicing window limits */
192 /* The indices read of arrays VWIN, INBUF, LPBUF, and BUFLIM */
193 /* are the same as those read by subroutine VPARMS. */
194 /* INBUF - Input speech buffer */
195 /* LPBUF - Low-pass filtered speech buffer */
196 /* BUFLIM - INBUF and LPBUF limits */
197 /* HALF - Present analysis half frame number */
198 /* MINAMD - Minimum value of the AMDF */
199 /* MAXAMD - Maximum value of the AMDF */
200 /* MINTAU - Pointer to the lag of the minimum AMDF value */
201 /* IVRC(2) - Inverse filter's RC's */
202 /* Only index 2 of array IVRC read under normal operation. */
203 /* (Index 1 is also read when debugging is turned on.) */
204 /* OBOUND - Onset boundary descriptions */
205 /* Indices 1 through 3 read if (HALF .NE. 1), otherwise untouched.
206 */
207 /* AF - The analysis frame number */
208 /* Output: */
209 /* VOIBUF(2,0:AF) - Buffer of voicing decisions */
210 /* Index (HALF,3) written. */
211 /* If (HALF .EQ. 1), skip down to "Read (HALF,3)" below. */
212 /* Indices (1,2), (2,1), (1,2), and (2,2) read. */
213 /* One of the following is then done: */
214 /* read (1,3) and possibly write (1,2) */
215 /* read (1,3) and write (1,2) or (2,2) */
216 /* write (2,1) */
217 /* write (2,1) or (1,2) */
218 /* read (1,0) and (1,3) and then write (2,2) or (1,1) */
219 /* no reads or writes on VOIBUF */
220 /* Finally, read (HALF,3) */
221 /* Internal: */
222 /* QS - Ratio of preemphasized to full-band energies */
223 /* RC1 - First reflection coefficient */
224 /* AR_B - Product of the causal forward and reverse pitch prediction gain
225 s*/
226 /* AR_F - Product of the noncausal forward and rev. pitch prediction gain
227 s*/
228 /* ZC - Zero crossing rate */
229 /* DITHER - Zero crossing threshold level */
230 /* MAXMIN - AMDF's 1 octave windowed maximum-to-minimum ratio */
231 /* MINPTR - Location of minimum AMDF value */
232 /* NVDC - Number of elements in each VDC vector */
233 /* NVDCL - Number of VDC vectors */
234 /* VDCL - SNR values corresponding to the set of VDC's */
235 /* VDC - 2-D voicing decision coefficient vector */
236 /* VALUE(9) - Voicing Parameters */
237 /* VOICE(2,3)- History of LDA results */
238 /* On every call when (HALF .EQ. 1), VOICE(*,I+1) is */
239 /* shifted back to VOICE(*,I), for I=1,2. */
240 /* VOICE(HALF,3) is written on every call. */
241 /* Depending on several conditions, one or more of */
242 /* (1,1), (1,2), (2,1), and (2,2) might then be read. */
243 /* LBE - Ratio of low-band instantaneous to average energies */
244 /* FBE - Ratio of full-band instantaneous to average energies */
245 /* LBVE - Low band voiced energy */
246 /* LBUE - Low band unvoiced energy */
247 /* FBVE - Full band voiced energy */
248 /* FBUE - Full band unvoiced energy */
249 /* OFBUE - Previous full-band unvoiced energy */
250 /* OLBUE - Previous low-band unvoiced energy */
251 /* REF - Reference energy for initialization and DITHER threshold */
252 /* SNR - Estimate of signal-to-noise ratio */
253 /* SNR2 - Estimate of low-band signal-to-noise ratio */
254 /* SNRL - SNR level number */
255 /* OT - Onset transition present */
256 /* VSTATE - Decimal interpretation of binary voicing classifications */
257 /* FIRST - First call flag */
258
259 /* This subroutine maintains local state from one call to the next. If */
260 /* you want to switch to using a new audio stream for this filter, or */
261 /* reinitialize its state for any other reason, call the ENTRY */
262 /* INITVOICIN. */
263
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)264 /* Subroutine */ int voicin_(integer *vwin, real *inbuf, real *
265 lpbuf, integer *buflim, integer *half, real *minamd, real *maxamd,
266 integer *mintau, real *ivrc, integer *obound, integer *voibuf,
267 integer *af, struct lpc10_encoder_state *st)
268 {
269 /* Initialized data */
270
271 real *dither;
272 static real vdc[100] /* was [10][10] */ = { 0.f,1714.f,-110.f,
273 334.f,-4096.f,-654.f,3752.f,3769.f,0.f,1181.f,0.f,874.f,-97.f,
274 300.f,-4096.f,-1021.f,2451.f,2527.f,0.f,-500.f,0.f,510.f,-70.f,
275 250.f,-4096.f,-1270.f,2194.f,2491.f,0.f,-1500.f,0.f,500.f,-10.f,
276 200.f,-4096.f,-1300.f,2e3f,2e3f,0.f,-2e3f,0.f,500.f,0.f,0.f,
277 -4096.f,-1300.f,2e3f,2e3f,0.f,-2500.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,
278 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,
279 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,
280 0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f,0.f };
281 static integer nvdcl = 5;
282 static real vdcl[10] = { 600.f,450.f,300.f,200.f,0.f,0.f,0.f,0.f,0.f,0.f }
283 ;
284
285 /* System generated locals */
286 integer inbuf_offset, lpbuf_offset, i__1, i__2;
287 real r__1, r__2;
288
289 /* Builtin functions */
290 integer i_nint(real *);
291 double sqrt(doublereal);
292
293 /* Local variables */
294 real ar_b__, ar_f__;
295 integer *lbve, *lbue, *fbve, *fbue;
296 integer snrl, i__;
297 integer *ofbue, *sfbue;
298 real *voice;
299 integer *olbue, *slbue;
300 real value[9];
301 integer zc;
302 logical ot;
303 real qs;
304 real *maxmin;
305 integer vstate;
306 real rc1;
307 extern /* Subroutine */ int vparms_(integer *, real *, real *, integer *,
308 integer *, real *, integer *, integer *, integer *, integer *,
309 real *, real *, real *, real *);
310 integer fbe, lbe;
311 real *snr;
312 real snr2;
313
314 /* Global Variables: */
315 /* Arguments */
316 /* $Log$
317 /* Revision 1.2 2006/08/01 13:06:51 rjongbloed
318 /* Added a raft of unvalidated audio codecs from OpenH323 tree
319 /*
320 /* Revision 1.1.2.1 2006/07/22 14:03:19 rjongbloed
321 /* Added more plug ins
322 /*
323 /* Revision 1.1.2.1 2006/05/08 13:49:58 rjongbloed
324 /* Imported all the audio codec plug ins from OpenH323
325 /*
326 /* Revision 1.1 2004/05/04 11:16:43 csoutheren
327 /* Initial version
328 /*
329 /* Revision 1.2 2002/02/15 03:57:55 yurik
330 /* Warnings removed during compilation, patch courtesy of Jehan Bing, jehan@bravobrava.com
331 /*
332 /* Revision 1.1 2000/06/05 04:45:12 robertj
333 /* Added LPC-10 2400bps codec
334 /*
335 * Revision 1.2 1996/08/20 20:45:00 jaf
336 * Removed all static local variables that were SAVE'd in the Fortran
337 * code, and put them in struct lpc10_encoder_state that is passed as an
338 * argument.
339 *
340 * Removed init function, since all initialization is now done in
341 * init_lpc10_encoder_state().
342 *
343 * Revision 1.1 1996/08/19 22:30:14 jaf
344 * Initial revision
345 * */
346 /* Revision 1.3 1996/03/29 22:05:55 jaf */
347 /* Commented out the common block variables that are not needed by the */
348 /* embedded version. */
349
350 /* Revision 1.2 1996/03/26 19:34:50 jaf */
351 /* Added comments indicating which constants are not needed in an */
352 /* application that uses the LPC-10 coder. */
353
354 /* Revision 1.1 1996/02/07 14:44:09 jaf */
355 /* Initial revision */
356
357 /* LPC Processing control variables: */
358
359 /* *** Read-only: initialized in setup */
360
361 /* Files for Speech, Parameter, and Bitstream Input & Output, */
362 /* and message and debug outputs. */
363
364 /* Here are the only files which use these variables: */
365
366 /* lpcsim.f setup.f trans.f error.f vqsetup.f */
367
368 /* Many files which use fdebug are not listed, since it is only used in */
369 /* those other files conditionally, to print trace statements. */
370 /* integer fsi, fso, fpi, fpo, fbi, fbo, pbin, fmsg, fdebug */
371 /* LPC order, Frame size, Quantization rate, Bits per frame, */
372 /* Error correction */
373 /* Subroutine SETUP is the only place where order is assigned a value, */
374 /* and that value is 10. It could increase efficiency 1% or so to */
375 /* declare order as a constant (i.e., a Fortran PARAMETER) instead of as
376 */
377 /* a variable in a COMMON block, since it is used in many places in the */
378 /* core of the coding and decoding routines. Actually, I take that back.
379 */
380 /* At least when compiling with f2c, the upper bound of DO loops is */
381 /* stored in a local variable before the DO loop begins, and then that is
382 */
383 /* compared against on each iteration. */
384 /* Similarly for lframe, which is given a value of MAXFRM in SETUP. */
385 /* Similarly for quant, which is given a value of 2400 in SETUP. quant */
386 /* is used in only a few places, and never in the core coding and */
387 /* decoding routines, so it could be eliminated entirely. */
388 /* nbits is similar to quant, and is given a value of 54 in SETUP. */
389 /* corrp is given a value of .TRUE. in SETUP, and is only used in the */
390 /* subroutines ENCODE and DECODE. It doesn't affect the speed of the */
391 /* coder significantly whether it is .TRUE. or .FALSE., or whether it is
392 */
393 /* a constant or a variable, since it is only examined once per frame. */
394 /* Leaving it as a variable that is set to .TRUE. seems like a good */
395 /* idea, since it does enable some error-correction capability for */
396 /* unvoiced frames, with no change in the coding rate, and no noticeable
397 */
398 /* quality difference in the decoded speech. */
399 /* integer quant, nbits */
400 /* *** Read/write: variables for debugging, not needed for LPC algorithm
401 */
402
403 /* Current frame, Unstable frames, Output clip count, Max onset buffer,
404 */
405 /* Debug listing detail level, Line count on listing page */
406
407 /* nframe is not needed for an embedded LPC10 at all. */
408 /* nunsfm is initialized to 0 in SETUP, and incremented in subroutine */
409 /* ERROR, which is only called from RCCHK. When LPC10 is embedded into */
410 /* an application, I would recommend removing the call to ERROR in RCCHK,
411 */
412 /* and remove ERROR and nunsfm completely. */
413 /* iclip is initialized to 0 in SETUP, and incremented in entry SWRITE in
414 */
415 /* sread.f. When LPC10 is embedded into an application, one might want */
416 /* to cause it to be incremented in a routine that takes the output of */
417 /* SYNTHS and sends it to an audio device. It could be optionally */
418 /* displayed, for those that might want to know what it is. */
419 /* maxosp is never initialized to 0 in SETUP, although it probably should
420 */
421 /* be, and it is updated in subroutine ANALYS. I doubt that its value */
422 /* would be of much interest to an application in which LPC10 is */
423 /* embedded. */
424 /* listl and lincnt are not needed for an embedded LPC10 at all. */
425 /* integer nframe, nunsfm, iclip, maxosp, listl, lincnt */
426 /* common /contrl/ fsi, fso, fpi, fpo, fbi, fbo, pbin, fmsg, fdebug */
427 /* common /contrl/ quant, nbits */
428 /* common /contrl/ nframe, nunsfm, iclip, maxosp, listl, lincnt */
429 /* Parameters/constants */
430 /* Voicing coefficient and Linear Discriminant Analysis variables:
431 */
432 /* Max number of VDC's and VDC levels */
433 /* The following are not Fortran PARAMETER's, but they are */
434 /* initialized with DATA statements, and never modified. */
435 /* Actual number of VDC's and levels */
436 /* Local variables that need not be saved */
437 /* Note: */
438
439 /* VALUE(1) through VALUE(8) are assigned values, but VALUE(9) */
440 /* never is. Yet VALUE(9) is read in the loop that begins "DO I =
441 */
442 /* 1, 9" below. I believe that this doesn't cause any problems in
443 */
444 /* this subroutine, because all VDC(9,*) array elements are 0, and
445 */
446 /* this is what is multiplied by VALUE(9) in all cases. Still, it
447 */
448 /* would save a multiplication to change the loop to "DO I = 1, 8".
449 */
450 /* Local state */
451 /* WARNING! */
452
453 /* VOICE, SFBUE, and SLBUE should be saved from one invocation to */
454 /* the next, but they are never given an initial value. */
455
456 /* Does Fortran 77 specify some default initial value, like 0, or */
457 /* is it undefined? If it is undefined, then this code should be */
458 /* corrected to specify an initial value. */
459
460 /* For VOICE, note that it is "shifted" in the statement that */
461 /* begins "IF (HALF .EQ. 1) THEN" below. Also, uninitialized */
462 /* values in the VOICE array can only affect entries in the VOIBUF
463 */
464 /* array that are for the same frame, or for an older frame. Thus
465 */
466 /* the effects of uninitialized values in VOICE cannot linger on */
467 /* for more than 2 or 3 frame times. */
468
469 /* For SFBUE and SLBUE, the effects of uninitialized values can */
470 /* linger on for many frame times, because their previous values */
471 /* are exponentially decayed. Thus it is more important to choose
472 */
473 /* initial values for these variables. I would guess that a */
474 /* reasonable initial value for SFBUE is REF/16, the same as used */
475 /* for FBUE and OFBUE. Similarly, SLBUE can be initialized to */
476 /* REF/32, the same as for LBUE and OLBUE. */
477
478 /* These guessed initial values should be validated by re-running */
479 /* the modified program on some audio samples. */
480
481 /* Declare and initialize filters: */
482
483 dither = (&st->dither);
484 snr = (&st->snr);
485 maxmin = (&st->maxmin);
486 voice = (&st->voice[0]);
487 lbve = (&st->lbve);
488 lbue = (&st->lbue);
489 fbve = (&st->fbve);
490 fbue = (&st->fbue);
491 ofbue = (&st->ofbue);
492 olbue = (&st->olbue);
493 sfbue = (&st->sfbue);
494 slbue = (&st->slbue);
495
496 /* Parameter adjustments */
497 if (vwin) {
498 --vwin;
499 }
500 if (buflim) {
501 --buflim;
502 }
503 if (inbuf) {
504 inbuf_offset = buflim[1];
505 inbuf -= inbuf_offset;
506 }
507 if (lpbuf) {
508 lpbuf_offset = buflim[3];
509 lpbuf -= lpbuf_offset;
510 }
511 if (ivrc) {
512 --ivrc;
513 }
514 if (obound) {
515 --obound;
516 }
517 if (voibuf) {
518 --voibuf;
519 }
520
521 /* Function Body */
522
523 /* The following variables are saved from one invocation to the */
524 /* next, but are not initialized with DATA statements. This is */
525 /* acceptable, because FIRST is initialized ot .TRUE., and the */
526 /* first time that this subroutine is then called, they are all */
527 /* given initial values. */
528
529 /* SNR */
530 /* LBVE, LBUE, FBVE, FBUE, OFBUE, OLBUE */
531
532 /* MAXMIN is initialized on the first call, assuming that HALF */
533 /* .EQ. 1 on first call. This is how ANALYS calls this subroutine.
534 */
535
536 /* Voicing Decision Parameter vector (* denotes zero coefficient): */
537
538 /* * MAXMIN */
539 /* LBE/LBVE */
540 /* ZC */
541 /* RC1 */
542 /* QS */
543 /* IVRC2 */
544 /* aR_B */
545 /* aR_F */
546 /* * LOG(LBE/LBVE) */
547 /* Define 2-D voicing decision coefficient vector according to the voicin
548 g*/
549 /* parameter order above. Each row (VDC vector) is optimized for a speci
550 fic*/
551 /* SNR. The last element of the vector is the constant. */
552 /* E ZC RC1 Qs IVRC2 aRb aRf c */
553
554 /* The VOICE array contains the result of the linear discriminant functio
555 n*/
556 /* (analog values). The VOIBUF array contains the hard-limited binary
557 */
558 /* voicing decisions. The VOICE and VOIBUF arrays, according to FORTRAN
559 */
560 /* memory allocation, are addressed as: */
561
562 /* (half-frame number, future-frame number) */
563
564 /* | Past | Present | Future1 | Future2 | */
565 /* | 1,0 | 2,0 | 1,1 | 2,1 | 1,2 | 2,2 | 1,3 | 2,3 | ---> time */
566
567 /* Update linear discriminant function history each frame: */
568 if (*half == 1) {
569 voice[0] = voice[2];
570 voice[1] = voice[3];
571 voice[2] = voice[4];
572 voice[3] = voice[5];
573 *maxmin = *maxamd / max(*minamd,1.f);
574 }
575 /* Calculate voicing parameters twice per frame: */
576 vparms_(&vwin[1], &inbuf[inbuf_offset], &lpbuf[lpbuf_offset], &buflim[1],
577 half, dither, mintau, &zc, &lbe, &fbe, &qs, &rc1, &ar_b__, &
578 ar_f__);
579 /* Estimate signal-to-noise ratio to select the appropriate VDC vector.
580 */
581 /* The SNR is estimated as the running average of the ratio of the */
582 /* running average full-band voiced energy to the running average */
583 /* full-band unvoiced energy. SNR filter has gain of 63. */
584 r__1 = (*snr + *fbve / (real) max(*fbue,1)) * 63 / 64.f;
585 *snr = (real) i_nint(&r__1);
586 snr2 = *snr * *fbue / max(*lbue,1);
587 /* Quantize SNR to SNRL according to VDCL thresholds. */
588 snrl = 1;
589 i__1 = nvdcl - 1;
590 for (snrl = 1; snrl <= i__1; ++snrl) {
591 if (snr2 > vdcl[snrl - 1]) {
592 goto L69;
593 }
594 }
595 /* (Note: SNRL = NVDCL here) */
596 L69:
597 /* Linear discriminant voicing parameters: */
598 value[0] = *maxmin;
599 value[1] = (real) lbe / max(*lbve,1);
600 value[2] = (real) zc;
601 value[3] = rc1;
602 value[4] = qs;
603 value[5] = ivrc[2];
604 value[6] = ar_b__;
605 value[7] = ar_f__;
606 /* Evaluation of linear discriminant function: */
607 voice[*half + 3] = vdc[snrl * 10 - 1];
608 for (i__ = 1; i__ <= 8; ++i__) {
609 voice[*half + 3] += vdc[i__ + snrl * 10 - 11] * value[i__ - 1];
610 }
611 /* Classify as voiced if discriminant > 0, otherwise unvoiced */
612 /* Voicing decision for current half-frame: 1 = Voiced; 0 = Unvoiced */
613 if (voice[*half + 3] > 0.f) {
614 voibuf[*half + 6] = 1;
615 } else {
616 voibuf[*half + 6] = 0;
617 }
618 /* Skip voicing decision smoothing in first half-frame: */
619 /* Give a value to VSTATE, so that trace statements below will print
620 */
621 /* a consistent value from one call to the next when HALF .EQ. 1. */
622 /* The value of VSTATE is not used for any other purpose when this is
623 */
624 /* true. */
625 vstate = -1;
626 if (*half == 1) {
627 goto L99;
628 }
629 /* Voicing decision smoothing rules (override of linear combination): */
630
631 /* Unvoiced half-frames: At least two in a row. */
632 /* -------------------- */
633
634 /* Voiced half-frames: At least two in a row in one frame. */
635 /* ------------------- Otherwise at least three in a row. */
636 /* (Due to the way transition frames are encoded) */
637
638 /* In many cases, the discriminant function determines how to smooth. */
639 /* In the following chart, the decisions marked with a * may be overridden
640 .*/
641
642 /* Voicing override of transitions at onsets: */
643 /* If a V/UV or UV/V voicing decision transition occurs within one-half
644 */
645 /* frame of an onset bounding a voicing window, then the transition is */
646 /* moved to occur at the onset. */
647
648 /* P 1F */
649 /* ----- ----- */
650 /* 0 0 0 0 */
651 /* 0 0 0* 1 (If there is an onset there) */
652 /* 0 0 1* 0* (Based on 2F and discriminant distance) */
653 /* 0 0 1 1 */
654 /* 0 1* 0 0 (Always) */
655 /* 0 1* 0* 1 (Based on discriminant distance) */
656 /* 0* 1 1 0* (Based on past, 2F, and discriminant distance) */
657 /* 0 1* 1 1 (If there is an onset there) */
658 /* 1 0* 0 0 (If there is an onset there) */
659 /* 1 0 0 1 */
660 /* 1 0* 1* 0 (Based on discriminant distance) */
661 /* 1 0* 1 1 (Always) */
662 /* 1 1 0 0 */
663 /* 1 1 0* 1* (Based on 2F and discriminant distance) */
664 /* 1 1 1* 0 (If there is an onset there) */
665 /* 1 1 1 1 */
666
667 /* Determine if there is an onset transition between P and 1F. */
668 /* OT (Onset Transition) is true if there is an onset between */
669 /* P and 1F but not after 1F. */
670 ot = ((obound[1] & 2) != 0 || obound[2] == 1) && (obound[3] & 1) == 0;
671 /* Multi-way dispatch on voicing decision history: */
672 vstate = (voibuf[3] << 3) + (voibuf[4] << 2) + (voibuf[5] << 1) + voibuf[
673 6];
674 switch (vstate + 1) {
675 case 1: goto L99;
676 case 2: goto L1;
677 case 3: goto L2;
678 case 4: goto L99;
679 case 5: goto L4;
680 case 6: goto L5;
681 case 7: goto L6;
682 case 8: goto L7;
683 case 9: goto L8;
684 case 10: goto L99;
685 case 11: goto L10;
686 case 12: goto L11;
687 case 13: goto L99;
688 case 14: goto L13;
689 case 15: goto L14;
690 case 16: goto L99;
691 }
692 L1:
693 if (ot && voibuf[7] == 1) {
694 voibuf[5] = 1;
695 }
696 goto L99;
697 L2:
698 if (voibuf[7] == 0 || voice[2] < -voice[3]) {
699 voibuf[5] = 0;
700 } else {
701 voibuf[6] = 1;
702 }
703 goto L99;
704 L4:
705 voibuf[4] = 0;
706 goto L99;
707 L5:
708 if (voice[1] < -voice[2]) {
709 voibuf[4] = 0;
710 } else {
711 voibuf[5] = 1;
712 }
713 goto L99;
714 /* VOIBUF(2,0) must be 0 */
715 L6:
716 if (voibuf[1] == 1 || voibuf[7] == 1 || voice[3] > voice[0]) {
717 voibuf[6] = 1;
718 } else {
719 voibuf[3] = 1;
720 }
721 goto L99;
722 L7:
723 if (ot) {
724 voibuf[4] = 0;
725 }
726 goto L99;
727 L8:
728 if (ot) {
729 voibuf[4] = 1;
730 }
731 goto L99;
732 L10:
733 if (voice[2] < -voice[1]) {
734 voibuf[5] = 0;
735 } else {
736 voibuf[4] = 1;
737 }
738 goto L99;
739 L11:
740 voibuf[4] = 1;
741 goto L99;
742 L13:
743 if (voibuf[7] == 0 && voice[3] < -voice[2]) {
744 voibuf[6] = 0;
745 } else {
746 voibuf[5] = 1;
747 }
748 goto L99;
749 L14:
750 if (ot && voibuf[7] == 0) {
751 voibuf[5] = 0;
752 }
753 /* GOTO 99 */
754 L99:
755 /* Now update parameters: */
756 /* ---------------------- */
757
758 /* During unvoiced half-frames, update the low band and full band unvoice
759 d*/
760 /* energy estimates (LBUE and FBUE) and also the zero crossing */
761 /* threshold (DITHER). (The input to the unvoiced energy filters is */
762 /* restricted to be less than 10dB above the previous inputs of the */
763 /* filters.) */
764 /* During voiced half-frames, update the low-pass (LBVE) and all-pass */
765 /* (FBVE) voiced energy estimates. */
766 if (voibuf[*half + 6] == 0) {
767 /* Computing MIN */
768 i__1 = fbe, i__2 = *ofbue * 3;
769 r__1 = (*sfbue * 63 + (min(i__1,i__2) << 3)) / 64.f;
770 *sfbue = i_nint(&r__1);
771 *fbue = *sfbue / 8;
772 *ofbue = fbe;
773 /* Computing MIN */
774 i__1 = lbe, i__2 = *olbue * 3;
775 r__1 = (*slbue * 63 + (min(i__1,i__2) << 3)) / 64.f;
776 *slbue = i_nint(&r__1);
777 *lbue = *slbue / 8;
778 *olbue = lbe;
779 } else {
780 r__1 = (*lbve * 63 + lbe) / 64.f;
781 *lbve = i_nint(&r__1);
782 r__1 = (*fbve * 63 + fbe) / 64.f;
783 *fbve = i_nint(&r__1);
784 }
785 /* Set dither threshold to yield proper zero crossing rates in the */
786 /* presence of low frequency noise and low level signal input. */
787 /* NOTE: The divisor is a function of REF, the expected energies. */
788 /* Computing MIN */
789 /* Computing MAX */
790 r__2 = (real)(sqrt((real) (*lbue * *lbve)) * 64 / 3000);
791 r__1 = max(r__2,1.f);
792 *dither = min(r__1,20.f);
793 /* Voicing decisions are returned in VOIBUF. */
794 return 0;
795 } /* voicin_ */
796