"Fossies" - the Fresh Open Source Software Archive 
1 /*************************************************************************/
2 /* */
3 /* Centre for Speech Technology Research */
4 /* University of Edinburgh, UK */
5 /* Copyright (c) 1995,1996 */
6 /* All Rights Reserved. */
7 /* */
8 /* Permission is hereby granted, free of charge, to use and distribute */
9 /* this software and its documentation without restriction, including */
10 /* without limitation the rights to use, copy, modify, merge, publish, */
11 /* distribute, sublicense, and/or sell copies of this work, and to */
12 /* permit persons to whom this work is furnished to do so, subject to */
13 /* the following conditions: */
14 /* 1. The code must retain the above copyright notice, this list of */
15 /* conditions and the following disclaimer. */
16 /* 2. Any modifications must be clearly marked as such. */
17 /* 3. Original authors' names are not deleted. */
18 /* 4. The authors' names are not used to endorse or promote products */
19 /* derived from this software without specific prior written */
20 /* permission. */
21 /* */
22 /* THE UNIVERSITY OF EDINBURGH AND THE CONTRIBUTORS TO THIS WORK */
23 /* DISCLAIM ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING */
24 /* ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO EVENT */
25 /* SHALL THE UNIVERSITY OF EDINBURGH NOR THE CONTRIBUTORS BE LIABLE */
26 /* FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES */
27 /* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN */
28 /* AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, */
29 /* ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF */
30 /* THIS SOFTWARE. */
31 /* */
32 /*************************************************************************/
33 /* Author : Paul Taylor */
34 /* Date : April 1994 */
35 /*************************************************************************/
36
37 #include "EST_speech_class.h"
38 #include "sigpr/EST_sigpr_utt.h"
39 #include "sigpr/EST_filter.h"
40 #include "srpd.h"
41 #include "EST_error.h"
42 #include "EST_string_aux.h"
43
44 int read_next_wave_segment (EST_Wave &sig, struct Srpd_Op *paras,
45 SEGMENT_ *p_seg);
46
47 static void srpd(EST_Wave &sig, EST_Track &fz, Srpd_Op &srpd_op, int resize);
48 static struct Srpd_Op *default_srpd_op(struct Srpd_Op *srpd);
49 static void parse_srpd_list(EST_Features &a_list, struct Srpd_Op *srpd);
50
51 void pda(EST_Wave &sig, EST_Track &fz, EST_Features &op, EST_String method)
52 {
53 if (method == "")
54 {
55 if (op.present("pda_method"))
56 method = op.S("pda_method");
57 }
58 if (method == "")
59 srpd(sig, fz, op);
60 else if (method == "srpd")
61 srpd(sig, fz, op);
62 else
63 EST_error("Unknown pda %s\n", (const char *)method);
64 }
65
66 void icda(EST_Wave &sig, EST_Track &fz, EST_Track &speech, EST_Features &op,
67 EST_String method)
68 { // intonation contour detection algorithm
69 EST_Track raw_fz;
70 if (method == "")
71 {
72 if (op.present("pda_method"))
73 method = op.S("pda_method");
74 }
75 if (method == "")
76 srpd(sig, raw_fz, op);
77 else if (method == "srpd")
78 srpd(sig, raw_fz, op);
79 else
80 EST_error("Unknown pda %s\n", (const char *)method);
81
82 smooth_phrase(raw_fz, speech, op, fz);
83 }
84
85 void srpd(EST_Wave &sig, EST_Track &fz, EST_Features &op)
86 {
87 Srpd_Op srpd_op;
88
89 default_srpd_op(&srpd_op); // default values
90 parse_srpd_list(op, &srpd_op); // override with options
91
92 if (op.I("do_low_pass",0))
93 FIRlowpass_filter(sig, op.I("lpf_cutoff"),op.I("lpf_order"));
94
95 srpd(sig, fz, srpd_op, op.I("srpd_resize", 0));
96 }
97
98 /*void do_srpd_fz(EST_Wave &sig, EST_Track &fz)
99 {
100 Srpd_Op srpd_op;
101 default_srpd_op(&srpd_op);
102 srpd(sig, fz, srpd_op, 1);
103 }
104 */
105
106 void srpd(EST_Wave &sig, EST_Track &fz, Srpd_Op &srpd_op, int resize)
107 {
108 int i, rns, tracklen, j = 0;
109 SEGMENT_ segment;
110 CROSS_CORR_ cc;
111 STATUS_ pda_status, held_status;
112 srpd_op.sample_freq = sig.sample_rate();
113 #if 0
114 float min, max;
115 min = srpd_op.min_pitch; // must store as set up routines corrupt
116 max = srpd_op.max_pitch;
117 #endif
118
119 initialise_structures (&srpd_op, &segment, &cc);
120 initialise_status (&srpd_op, &pda_status);
121 initialise_status (&srpd_op, &held_status);
122
123 tracklen = (sig.num_samples() - segment.length) / segment.shift + 1;
124
125 if (resize)
126 {
127 fz.set_equal_space(true);
128 fz.resize(tracklen, 1);
129 fz.set_channel_name("F0", 0);
130 fz.fill_time(srpd_op.shift/1000);
131 }
132
133 if (!fz.equal_space())
134 EST_error("Pitch tracking algorithm must have equal spaced track\n");
135
136 while ((rns = read_next_wave_segment (sig, &srpd_op, &segment)) != 0)
137 {
138 if (rns == 2)
139 {
140 for (i = 0; i < cc.size; cc.coeff[i++] = 0.0);
141 initialise_status (&srpd_op, &pda_status);
142 }
143 else
144 super_resolution_pda (&srpd_op, segment, &cc, &pda_status);
145 if (pda_status.s_h == HOLD)
146 {
147 held_status.pitch_freq = pda_status.pitch_freq;
148 held_status.v_uv = VOICED;
149 held_status.s_h = HELD;
150 held_status.cc_max = pda_status.cc_max;
151 held_status.threshold = pda_status.threshold;
152 continue;
153 }
154 if (held_status.s_h == HELD)
155 {
156 if (pda_status.pitch_freq == BREAK_NUMBER)
157 {
158 held_status.pitch_freq = BREAK_NUMBER;
159 held_status.v_uv = UNVOICED;
160 }
161 held_status.s_h = SENT;
162 if (held_status.v_uv != VOICED)
163 fz.set_break(j);
164 fz.a(j++) = held_status.pitch_freq;
165 // printf( "track set: %d (of %d) to %f\n", j-1, fz.length(), held_status.pitch_freq );
166 }
167 if (pda_status.v_uv != VOICED)
168 fz.set_break(j);
169 fz.a(j++) = pda_status.pitch_freq;
170 //printf( "track set: %d (of %d) to %f\n", j-1, fz.length(), pda_status.pitch_freq );
171 }
172 if (held_status.s_h == HELD)
173 {
174 held_status.pitch_freq = BREAK_NUMBER;
175 held_status.v_uv = UNVOICED;
176 fz.set_break(j);
177 fz.a(j++) = held_status.pitch_freq;
178 }
179 end_structure_use (&segment, &cc);
180 }
181
182 static struct Srpd_Op *default_srpd_op(struct Srpd_Op *srpd)
183 {
184 srpd->L = DEFAULT_DECIMATION;
185 srpd->min_pitch = DEFAULT_MIN_PITCH;
186 srpd->max_pitch = DEFAULT_MAX_PITCH;
187 srpd->shift = DEFAULT_SHIFT;
188 srpd->length = DEFAULT_LENGTH;
189 srpd->Tsilent = DEFAULT_TSILENT;
190 srpd->Tmin = DEFAULT_TMIN;
191 srpd->Tmax_ratio = DEFAULT_TMAX_RATIO;
192 srpd->Thigh = DEFAULT_THIGH;
193 srpd->Tdh = DEFAULT_TDH;
194 srpd->make_ascii = 0;
195 srpd->peak_tracking = 0;
196 srpd->sample_freq = DEFAULT_SF;
197 /* p_par->Nmax and p_par->Nmin cannot be initialised */
198 return(srpd);
199 }
200
201 static void parse_srpd_list(EST_Features &al, struct Srpd_Op *srpd)
202 {
203 if (al.present("decimation"))
204 srpd->L = al.I("decimation");
205 if (al.present("min_pitch"))
206 srpd->min_pitch = al.F("min_pitch");
207 if (al.present("max_pitch"))
208 srpd->max_pitch = al.F("max_pitch");
209 if (al.present("pda_frame_shift"))
210 srpd->shift = al.F("pda_frame_shift") * 1000.0;
211 if (al.present("pda_frame_length"))
212 srpd->length = al.F("pda_frame_length") * 1000.0;
213 if (al.present("noise_floor"))
214 srpd->Tsilent = al.I("noise_floor");
215 if (al.present("v2uv_coeff_thresh"))
216 srpd->Thigh = al.F("v2uv_coef_thresh");
217 if (al.present("min_v2uv_coef_thresh"))
218 srpd->Tmin = al.F("min_v2uv_coef_thresh");
219 if (al.present("v2uv_coef_thresh_ratio"))
220 srpd->Tmax_ratio = al.F("v2uv_coef_thresh_ratio");
221 if (al.present("anti_doubling_thresh"))
222 srpd->Tdh = al.F("anti_doubling_thresh");
223 if (al.present("peak_tracking"))
224 srpd->peak_tracking = al.I("peak_tracking");
225 if (al.present("sample_frequency"))
226 srpd->sample_freq = al.I("sample_frequency");
227 }
228
229 void default_pda_options(EST_Features &al)
230 {
231 al.set("min_pitch", "40.0");
232 al.set("max_pitch", "400.0");
233 al.set("pda_frame_shift", "0.005");
234 al.set("pda_frame_length", DEFAULT_LENGTH / 1000.0);
235 al.set("lpf_cutoff", "600");
236 al.set("lpf_order", "49");
237 al.set("f0_file_type", "esps");
238 al.set("decimation", DEFAULT_DECIMATION);
239 al.set("noise_floor", DEFAULT_TSILENT);
240 al.set("min_v2uv_coef_thresh", DEFAULT_TMIN);
241 al.set("v2uv_coef_thresh_ratio", DEFAULT_TMAX_RATIO);
242 al.set("v2uv_coef_thresh", DEFAULT_THIGH);
243 al.set("anti_doubling_thresh", DEFAULT_TDH);
244 al.set("peak_tracking", 0);
245 }
246
247 EST_String options_pda_general(void)
248 {
249 // The standard waveform input options
250 return
251 EST_String("")+
252 "-L Perform low pass filtering on input. This option should always \n"
253 " be used in normal processing as it usually increases \n"
254 " performance considerably\n\n"
255 "-P perform peak tracking\n\n"
256 "-fmin <float> miniumum F0 value. Sets the minimum allowed F0 in \n"
257 " output track. Default is "+ftoString(DEFAULT_MIN_PITCH)+".\n "
258 " Changing this to suit the speaker usually increases \n"
259 " performance. Typical recommended values are 60-90Hz for\n"
260 " males and 120-150Hz for females\n\n"
261 "-fmax <float> maxiumum F0 value. Sets the maximum allowed F0 in \n"
262 " output track. Default is "+ftoString(DEFAULT_MAX_PITCH)+". \n"
263 " Changing this to suit the speaker usually increases \n"
264 " performance. Typical recommended values are 200Hz for \n"
265 " males and 300-400Hz for females\n\n"
266 "-shift <float> frame spacing in seconds for fixed frame analysis. \n"
267 " This doesn't have to be the same as the output file spacing - \n"
268 " the -S option can be used to resample the track before saving \n"
269 " default: "+ftoString(DEFAULT_SHIFT/1000.0) +"\n\n"
270 "-length <float> analysis frame length in seconds.\n"
271 " default: "+ftoString(DEFAULT_LENGTH/1000.0) +"\n\n"
272 "-lpfilter <int> Low pass filter, with cutoff frequency in Hz \n"
273 " Filtering is performed by a FIR filter which is built at run \n"
274 " time. The order of the filter can be given by -forder. The \n"
275 " default value is 199\n\n"
276 "-forder <int> Order of FIR filter used for lpfilter and \n"
277 " hpfilter. This must be ODD. Sensible values range \n"
278 " from 19 (quick but with a shallow rolloff) to 199 \n"
279 " (slow but with a steep rolloff). The default is 199.\n\n";
280 }
281
282 EST_String options_pda_srpd(void)
283 {
284 // The standard waveform input options
285 return
286 EST_String("")+
287 "-d <float> decimation factor\n"
288 " set down-sampling for quicker computation so that only one in \n"
289 " <parameter>decimation factor</parameter> samples are used in the first instance. \n"
290 " Must be in the range of one to ten inclusive. Default is four. \n"
291 " For data sampled at 10kHz, it is advised that a decimation \n"
292 " factor of two isselected.\n\n"
293
294 "-n <float> Inoise floor.\n"
295 " Set the maximum absolute signal amplitude that represents \n"
296 " silence to <parameter>Inoise floor</parameter>. If the absolute amplitude of \n"
297 " the first segment in a given frame is below this level at all \n"
298 " times, then the frame is classified as representing silence. \n"
299 " Must be a positive number. Default is 120 ADC units.\n\n"
300
301 "-H <float> unvoiced to voiced coeff threshold\n"
302 " set the correlation coefficient threshold which must be \n"
303 " exceeded in a transition from an unvoiced classified frame \n"
304 " of speech to a voiced frame as the unvoiced to voiced coeff \n"
305 " threshold. Must be in the range zero to one inclusive. \n"
306 " Default is 0.88.\n\n"
307
308 "-m <float> min voiced to unvoiced coeff threshold \n"
309 " set the minimum allowed correlation coefficient threshold \n"
310 " which must not be exceeded in a transition from a voiced \n"
311 " classified frame of speech to an unvoiced frame, as \n"
312 " <parameter>min voiced to unvoiced coeff threshold</parameter>. Must be in the \n"
313 " range zero to <parameter>unvoiced to voiced coeff threshold</parameter> \n"
314 " inclusive. Default is 0.75.\n\n"
315
316 "-R <float> voiced to unvoiced coeff threshold-ratio \n"
317 " set the scaling factor used in determining the correlation\n"
318 " coefficient threshold which must not be exceeded in a voiced \n"
319 " frame to unvoiced frame transition, as <parameter>voiced to unvoiced</parameter> \n"
320 " coeff threshold -ratio. The voiced to unvoiced coefficient \n"
321 " threshold is determined by multiplying this scaling factor \n"
322 " with the maximum cross-correlation coefficient of the \n"
323 " previously voiced frame. If this product is less than \n"
324 " <parameter>min voiced to unvoiced coeff threshold</parameter> then this is used \n"
325 " instead. Must be in the range zero to one inclusive. \n"
326 " Default is 0.85.\n\n"
327
328 "-t <float> anti pitch doubling/halving threshold\n"
329 " set the threshold used in eliminating (as far as possible) \n"
330 " pitch doubling and pitch halving errors as <parameter>anti pitch \n"
331 " double/halving threshold</parameter>. Must be in the range zero to \n"
332 " one inclusive. Default is 0.77.\n\n";
333 }
334
335
2.4-release_vs_2.5.0-release.