|
5
|
1 /*
|
|
|
2 * SpanDSP - a series of DSP components for telephony
|
|
|
3 *
|
|
|
4 * time_scale.c - Time scaling for linear speech data
|
|
|
5 *
|
|
|
6 * Written by Steve Underwood <steveu@coppice.org>
|
|
|
7 *
|
|
|
8 * Copyright (C) 2004 Steve Underwood
|
|
|
9 *
|
|
|
10 * All rights reserved.
|
|
|
11 *
|
|
|
12 * This program is free software; you can redistribute it and/or modify
|
|
|
13 * it under the terms of the GNU General Public License version 2, as
|
|
|
14 * published by the Free Software Foundation.
|
|
|
15 *
|
|
|
16 * This program is distributed in the hope that it will be useful,
|
|
|
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
19 * GNU General Public License for more details.
|
|
|
20 *
|
|
|
21 * You should have received a copy of the GNU General Public License
|
|
|
22 * along with this program; if not, write to the Free Software
|
|
|
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
|
|
|
24 *
|
|
|
25 * $Id: time_scale.c,v 1.15 2006/11/19 14:07:25 steveu Exp $
|
|
|
26 */
|
|
|
27
|
|
|
28 /*! \file */
|
|
|
29
|
|
|
30 #ifdef HAVE_CONFIG_H
|
|
|
31 #include <config.h>
|
|
|
32 #endif
|
|
|
33
|
|
|
34 #include <stdlib.h>
|
|
|
35 #include <stdio.h>
|
|
|
36 #include <inttypes.h>
|
|
|
37 #include <string.h>
|
|
|
38 #include <fcntl.h>
|
|
|
39 #include <time.h>
|
|
|
40 #include <limits.h>
|
|
|
41 #if defined(HAVE_TGMATH_H)
|
|
|
42 #include <tgmath.h>
|
|
|
43 #endif
|
|
|
44 #if defined(HAVE_MATH_H)
|
|
|
45 #include <math.h>
|
|
|
46 #endif
|
|
|
47
|
|
|
48 #include "spandsp/telephony.h"
|
|
|
49 #include "spandsp/time_scale.h"
|
|
|
50
|
|
|
51 /*
|
|
|
52 Time scaling for speech, based on the Pointer Interval Controlled
|
|
|
53 OverLap and Add (PICOLA) method, developed by Morita Naotaka.
|
|
|
54 */
|
|
|
55
|
|
|
56 static __inline__ int amdf_pitch(int min_pitch, int max_pitch, int16_t amp[], int len)
|
|
|
57 {
|
|
|
58 int i;
|
|
|
59 int j;
|
|
|
60 int acc;
|
|
|
61 int min_acc;
|
|
|
62 int pitch;
|
|
|
63
|
|
|
64 pitch = min_pitch;
|
|
|
65 min_acc = INT_MAX;
|
|
|
66 for (i = max_pitch; i <= min_pitch; i++)
|
|
|
67 {
|
|
|
68 acc = 0;
|
|
|
69 for (j = 0; j < len; j++)
|
|
|
70 acc += abs(amp[i + j] - amp[j]);
|
|
|
71 if (acc < min_acc)
|
|
|
72 {
|
|
|
73 min_acc = acc;
|
|
|
74 pitch = i;
|
|
|
75 }
|
|
|
76 }
|
|
|
77 return pitch;
|
|
|
78 }
|
|
|
79 /*- End of function --------------------------------------------------------*/
|
|
|
80
|
|
|
81 static __inline__ void overlap_add(int16_t amp1[], int16_t amp2[], int len)
|
|
|
82 {
|
|
|
83 int i;
|
|
|
84 double weight;
|
|
|
85 double step;
|
|
|
86
|
|
|
87 step = 1.0/len;
|
|
|
88 weight = 0.0;
|
|
|
89 for (i = 0; i < len; i++)
|
|
|
90 {
|
|
|
91 /* TODO: saturate */
|
|
|
92 amp2[i] = (int16_t) ((double) amp1[i]*(1.0 - weight) + (double) amp2[i]*weight);
|
|
|
93 weight += step;
|
|
|
94 }
|
|
|
95 }
|
|
|
96 /*- End of function --------------------------------------------------------*/
|
|
|
97
|
|
|
98 int time_scale_rate(time_scale_t *s, float rate)
|
|
|
99 {
|
|
|
100 if (rate <= 0.0)
|
|
|
101 return -1;
|
|
|
102 /*endif*/
|
|
|
103 if (rate >= 0.99 && rate <= 1.01)
|
|
|
104 {
|
|
|
105 /* Treat rate close to normal speed as exactly normal speed, and
|
|
|
106 avoid divide by zero, and other numerical problems. */
|
|
|
107 rate = 1.0;
|
|
|
108 }
|
|
|
109 else if (rate < 1.0)
|
|
|
110 {
|
|
|
111 s->rcomp = rate/(1.0 - rate);
|
|
|
112 }
|
|
|
113 else
|
|
|
114 {
|
|
|
115 s->rcomp = 1.0/(rate - 1.0);
|
|
|
116 }
|
|
|
117 /*endif*/
|
|
|
118 s->rate = rate;
|
|
|
119 return 0;
|
|
|
120 }
|
|
|
121 /*- End of function --------------------------------------------------------*/
|
|
|
122
|
|
|
123 int time_scale_init(time_scale_t *s, float rate)
|
|
|
124 {
|
|
|
125 if (time_scale_rate(s, rate))
|
|
|
126 return -1;
|
|
|
127 /*endif*/
|
|
|
128 s->rate_nudge = 0.0;
|
|
|
129 s->fill = 0;
|
|
|
130 s->lcp = 0;
|
|
|
131 return 0;
|
|
|
132 }
|
|
|
133 /*- End of function --------------------------------------------------------*/
|
|
|
134
|
|
|
135 int time_scale(time_scale_t *s, int16_t out[], int16_t in[], int len)
|
|
|
136 {
|
|
|
137 double lcpf;
|
|
|
138 int pitch;
|
|
|
139 int out_len;
|
|
|
140 int in_len;
|
|
|
141 int k;
|
|
|
142
|
|
|
143 out_len = 0;
|
|
|
144 in_len = 0;
|
|
|
145
|
|
|
146 /* Top up the buffer */
|
|
|
147 if (s->fill + len < TIME_SCALE_BUF_LEN)
|
|
|
148 {
|
|
|
149 /* Cannot continue without more samples */
|
|
|
150 memcpy(s->buf + s->fill, in, sizeof(int16_t)*len);
|
|
|
151 s->fill += len;
|
|
|
152 return out_len;
|
|
|
153 }
|
|
|
154 k = (TIME_SCALE_BUF_LEN - s->fill);
|
|
|
155 memcpy(s->buf + s->fill, in, sizeof(int16_t)*k);
|
|
|
156 in_len += k;
|
|
|
157 s->fill = TIME_SCALE_BUF_LEN;
|
|
|
158 while (s->fill == TIME_SCALE_BUF_LEN)
|
|
|
159 {
|
|
|
160 while (s->lcp >= TIME_SCALE_BUF_LEN)
|
|
|
161 {
|
|
|
162 memcpy(out + out_len, s->buf, sizeof(int16_t)*TIME_SCALE_BUF_LEN);
|
|
|
163 out_len += TIME_SCALE_BUF_LEN;
|
|
|
164 if (len - in_len < TIME_SCALE_BUF_LEN)
|
|
|
165 {
|
|
|
166 /* Cannot continue without more samples */
|
|
|
167 memcpy(s->buf, in + in_len, sizeof(int16_t)*(len - in_len));
|
|
|
168 s->fill = len - in_len;
|
|
|
169 s->lcp -= TIME_SCALE_BUF_LEN;
|
|
|
170 return out_len;
|
|
|
171 }
|
|
|
172 memcpy(s->buf, in + in_len, sizeof(int16_t)*TIME_SCALE_BUF_LEN);
|
|
|
173 in_len += TIME_SCALE_BUF_LEN;
|
|
|
174 s->lcp -= TIME_SCALE_BUF_LEN;
|
|
|
175 }
|
|
|
176 if (s->lcp > 0)
|
|
|
177 {
|
|
|
178 memcpy(out + out_len, s->buf, sizeof(int16_t)*s->lcp);
|
|
|
179 out_len += s->lcp;
|
|
|
180 memcpy(s->buf, s->buf + s->lcp, sizeof(int16_t)*(TIME_SCALE_BUF_LEN - s->lcp));
|
|
|
181 if (len - in_len < s->lcp)
|
|
|
182 {
|
|
|
183 /* Cannot continue without more samples */
|
|
|
184 memcpy(s->buf + (TIME_SCALE_BUF_LEN - s->lcp), in + in_len, sizeof(int16_t)*(len - in_len));
|
|
|
185 s->fill = TIME_SCALE_BUF_LEN - s->lcp + len - in_len;
|
|
|
186 s->lcp = 0;
|
|
|
187 return out_len;
|
|
|
188 }
|
|
|
189 memcpy(s->buf + (TIME_SCALE_BUF_LEN - s->lcp), in + in_len, sizeof(int16_t)*s->lcp);
|
|
|
190 in_len += s->lcp;
|
|
|
191 s->lcp = 0;
|
|
|
192 }
|
|
|
193 if (s->rate == 1.0)
|
|
|
194 {
|
|
|
195 s->lcp = 0x7FFFFFFF;
|
|
|
196 }
|
|
|
197 else
|
|
|
198 {
|
|
|
199 pitch = amdf_pitch(SAMPLE_RATE/TIME_SCALE_MIN_PITCH, SAMPLE_RATE/TIME_SCALE_MAX_PITCH, s->buf, SAMPLE_RATE/TIME_SCALE_MIN_PITCH);
|
|
|
200 lcpf = (double) pitch*s->rcomp;
|
|
|
201 /* Nudge around to compensate for fractional samples */
|
|
|
202 s->lcp = (int) lcpf;
|
|
|
203 /* Note that s->lcp and lcpf are not the same, as lcpf has a fractional part, and s->lcp doesn't */
|
|
|
204 s->rate_nudge += s->lcp - lcpf;
|
|
|
205 if (s->rate_nudge >= 0.5)
|
|
|
206 {
|
|
|
207 s->lcp--;
|
|
|
208 s->rate_nudge -= 1.0;
|
|
|
209 }
|
|
|
210 else if (s->rate_nudge <= -0.5)
|
|
|
211 {
|
|
|
212 s->lcp++;
|
|
|
213 s->rate_nudge += 1.0;
|
|
|
214 }
|
|
|
215 if (s->rate < 1.0)
|
|
|
216 {
|
|
|
217 /* Speed up - drop a chunk of data */
|
|
|
218 overlap_add(s->buf, s->buf + pitch, pitch);
|
|
|
219 memcpy(&s->buf[pitch], &s->buf[2*pitch], sizeof(int16_t)*(TIME_SCALE_BUF_LEN - 2*pitch));
|
|
|
220 if (len - in_len < pitch)
|
|
|
221 {
|
|
|
222 /* Cannot continue without more samples */
|
|
|
223 memcpy(s->buf + TIME_SCALE_BUF_LEN - pitch, in + in_len, sizeof(int16_t)*(len - in_len));
|
|
|
224 s->fill += (len - in_len - pitch);
|
|
|
225 return out_len;
|
|
|
226 }
|
|
|
227 memcpy(s->buf + TIME_SCALE_BUF_LEN - pitch, in + in_len, sizeof(int16_t)*pitch);
|
|
|
228 in_len += pitch;
|
|
|
229 }
|
|
|
230 else
|
|
|
231 {
|
|
|
232 /* Slow down - insert a chunk of data */
|
|
|
233 memcpy(out + out_len, s->buf, sizeof(int16_t)*pitch);
|
|
|
234 out_len += pitch;
|
|
|
235 overlap_add(s->buf + pitch, s->buf, pitch);
|
|
|
236 }
|
|
|
237 }
|
|
|
238 }
|
|
|
239 return out_len;
|
|
|
240 }
|
|
|
241 /*- End of function --------------------------------------------------------*/
|
|
|
242 /*- End of file ------------------------------------------------------------*/
|
