GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavcodec/truespeech.c Lines: 181 186 97.3 %
Date: 2020-09-25 23:16:12 Branches: 69 72 95.8 %

Line Branch Exec Source
1
/*
2
 * DSP Group TrueSpeech compatible decoder
3
 * Copyright (c) 2005 Konstantin Shishkov
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21
22
#include "libavutil/channel_layout.h"
23
#include "libavutil/intreadwrite.h"
24
#include "avcodec.h"
25
#include "bswapdsp.h"
26
#include "get_bits.h"
27
#include "internal.h"
28
29
#include "truespeech_data.h"
30
/**
31
 * @file
32
 * TrueSpeech decoder.
33
 */
34
35
/**
36
 * TrueSpeech decoder context
37
 */
38
typedef struct TSContext {
39
    BswapDSPContext bdsp;
40
    /* input data */
41
    DECLARE_ALIGNED(16, uint8_t, buffer)[32];
42
    int16_t vector[8];  ///< input vector: 5/5/4/4/4/3/3/3
43
    int offset1[2];     ///< 8-bit value, used in one copying offset
44
    int offset2[4];     ///< 7-bit value, encodes offsets for copying and for two-point filter
45
    int pulseoff[4];    ///< 4-bit offset of pulse values block
46
    int pulsepos[4];    ///< 27-bit variable, encodes 7 pulse positions
47
    int pulseval[4];    ///< 7x2-bit pulse values
48
    int flag;           ///< 1-bit flag, shows how to choose filters
49
    /* temporary data */
50
    int filtbuf[146];   // some big vector used for storing filters
51
    int prevfilt[8];    // filter from previous frame
52
    int16_t tmp1[8];    // coefficients for adding to out
53
    int16_t tmp2[8];    // coefficients for adding to out
54
    int16_t tmp3[8];    // coefficients for adding to out
55
    int16_t cvector[8]; // correlated input vector
56
    int filtval;        // gain value for one function
57
    int16_t newvec[60]; // tmp vector
58
    int16_t filters[32]; // filters for every subframe
59
} TSContext;
60
61
2
static av_cold int truespeech_decode_init(AVCodecContext * avctx)
62
{
63
2
    TSContext *c = avctx->priv_data;
64
65
2
    if (avctx->channels != 1) {
66
        avpriv_request_sample(avctx, "Channel count %d", avctx->channels);
67
        return AVERROR_PATCHWELCOME;
68
    }
69
70
2
    avctx->channel_layout = AV_CH_LAYOUT_MONO;
71
2
    avctx->sample_fmt     = AV_SAMPLE_FMT_S16;
72
73
2
    ff_bswapdsp_init(&c->bdsp);
74
75
2
    return 0;
76
}
77
78
1442
static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
79
{
80
    GetBitContext gb;
81
82
1442
    dec->bdsp.bswap_buf((uint32_t *) dec->buffer, (const uint32_t *) input, 8);
83
1442
    init_get_bits(&gb, dec->buffer, 32 * 8);
84
85
1442
    dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)];
86
1442
    dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)];
87
1442
    dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)];
88
1442
    dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)];
89
1442
    dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)];
90
1442
    dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)];
91
1442
    dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)];
92
1442
    dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)];
93
1442
    dec->flag      = get_bits1(&gb);
94
95
1442
    dec->offset1[0] = get_bits(&gb, 4) << 4;
96
1442
    dec->offset2[3] = get_bits(&gb, 7);
97
1442
    dec->offset2[2] = get_bits(&gb, 7);
98
1442
    dec->offset2[1] = get_bits(&gb, 7);
99
1442
    dec->offset2[0] = get_bits(&gb, 7);
100
101
1442
    dec->offset1[1]  = get_bits(&gb, 4);
102
1442
    dec->pulseval[1] = get_bits(&gb, 14);
103
1442
    dec->pulseval[0] = get_bits(&gb, 14);
104
105
1442
    dec->offset1[1] |= get_bits(&gb, 4) << 4;
106
1442
    dec->pulseval[3] = get_bits(&gb, 14);
107
1442
    dec->pulseval[2] = get_bits(&gb, 14);
108
109
1442
    dec->offset1[0] |= get_bits1(&gb);
110
1442
    dec->pulsepos[0] = get_bits_long(&gb, 27);
111
1442
    dec->pulseoff[0] = get_bits(&gb, 4);
112
113
1442
    dec->offset1[0] |= get_bits1(&gb) << 1;
114
1442
    dec->pulsepos[1] = get_bits_long(&gb, 27);
115
1442
    dec->pulseoff[1] = get_bits(&gb, 4);
116
117
1442
    dec->offset1[0] |= get_bits1(&gb) << 2;
118
1442
    dec->pulsepos[2] = get_bits_long(&gb, 27);
119
1442
    dec->pulseoff[2] = get_bits(&gb, 4);
120
121
1442
    dec->offset1[0] |= get_bits1(&gb) << 3;
122
1442
    dec->pulsepos[3] = get_bits_long(&gb, 27);
123
1442
    dec->pulseoff[3] = get_bits(&gb, 4);
124
1442
}
125
126
1442
static void truespeech_correlate_filter(TSContext *dec)
127
{
128
    int16_t tmp[8];
129
    int i, j;
130
131
12978
    for(i = 0; i < 8; i++){
132
11536
        if(i > 0){
133
10094
            memcpy(tmp, dec->cvector, i * sizeof(*tmp));
134
50470
            for(j = 0; j < i; j++)
135
40376
                dec->cvector[j] += (tmp[i - j - 1] * dec->vector[i] + 0x4000) >> 15;
136
        }
137
11536
        dec->cvector[i] = (8 - dec->vector[i]) >> 3;
138
    }
139
12978
    for(i = 0; i < 8; i++)
140
11536
        dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15;
141
142
1442
    dec->filtval = dec->vector[0];
143
1442
}
144
145
1442
static void truespeech_filters_merge(TSContext *dec)
146
{
147
    int i;
148
149
1442
    if(!dec->flag){
150
99
        for(i = 0; i < 8; i++){
151
88
            dec->filters[i + 0] = dec->prevfilt[i];
152
88
            dec->filters[i + 8] = dec->prevfilt[i];
153
        }
154
    }else{
155
12879
        for(i = 0; i < 8; i++){
156
11448
            dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15;
157
11448
            dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15;
158
        }
159
    }
160
12978
    for(i = 0; i < 8; i++){
161
11536
        dec->filters[i + 16] = dec->cvector[i];
162
11536
        dec->filters[i + 24] = dec->cvector[i];
163
    }
164
1442
}
165
166
5768
static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
167
{
168
    int16_t tmp[146 + 60], *ptr0, *ptr1;
169
    const int16_t *filter;
170
    int i, t, off;
171
172
5768
    t = dec->offset2[quart];
173
5768
    if(t == 127){
174
358
        memset(dec->newvec, 0, 60 * sizeof(*dec->newvec));
175
358
        return;
176
    }
177
795270
    for(i = 0; i < 146; i++)
178
789860
        tmp[i] = dec->filtbuf[i];
179
5410
    off = (t / 25) + dec->offset1[quart >> 1] + 18;
180
5410
    off = av_clip(off, 0, 145);
181
5410
    ptr0 = tmp + 145 - off;
182
5410
    ptr1 = tmp + 146;
183
5410
    filter = ts_order2_coeffs + (t % 25) * 2;
184
330010
    for(i = 0; i < 60; i++){
185
324600
        t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14;
186
324600
        ptr0++;
187
324600
        dec->newvec[i] = t;
188
324600
        ptr1[i] = t;
189
    }
190
}
191
192
5768
static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
193
{
194
    int16_t tmp[7];
195
    int i, j, t;
196
    const int16_t *ptr1;
197
    int16_t *ptr2;
198
    int coef;
199
200
5768
    memset(out, 0, 60 * sizeof(*out));
201
46144
    for(i = 0; i < 7; i++) {
202
40376
        t = dec->pulseval[quart] & 3;
203
40376
        dec->pulseval[quart] >>= 2;
204
40376
        tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t];
205
    }
206
207
5768
    coef = dec->pulsepos[quart] >> 15;
208
5768
    ptr1 = ts_pulse_values + 30;
209
5768
    ptr2 = tmp;
210

136138
    for(i = 0, j = 3; (i < 30) && (j > 0); i++){
211
130370
        t = *ptr1++;
212
130370
        if(coef >= t)
213
113066
            coef -= t;
214
        else{
215
17304
            out[i] = *ptr2++;
216
17304
            ptr1 += 30;
217
17304
            j--;
218
        }
219
    }
220
5768
    coef = dec->pulsepos[quart] & 0x7FFF;
221
5768
    ptr1 = ts_pulse_values;
222

141397
    for(i = 30, j = 4; (i < 60) && (j > 0); i++){
223
135629
        t = *ptr1++;
224
135629
        if(coef >= t)
225
112557
            coef -= t;
226
        else{
227
23072
            out[i] = *ptr2++;
228
23072
            ptr1 += 30;
229
23072
            j--;
230
        }
231
    }
232
233
5768
}
234
235
5768
static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
236
{
237
    int i;
238
239
5768
    memmove(dec->filtbuf, &dec->filtbuf[60], 86 * sizeof(*dec->filtbuf));
240
351848
    for(i = 0; i < 60; i++){
241
346080
        dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3);
242
346080
        out[i] += dec->newvec[i];
243
    }
244
5768
}
245
246
5768
static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
247
{
248
    int i,k;
249
    int t[8];
250
    int16_t *ptr0, *ptr1;
251
252
5768
    ptr0 = dec->tmp1;
253
5768
    ptr1 = dec->filters + quart * 8;
254
351848
    for(i = 0; i < 60; i++){
255
346080
        int sum = 0;
256
3114720
        for(k = 0; k < 8; k++)
257
2768640
            sum += ptr0[k] * (unsigned)ptr1[k];
258
346080
        sum = out[i] + ((int)(sum + 0x800U) >> 12);
259
346080
        out[i] = av_clip(sum, -0x7FFE, 0x7FFE);
260
2768640
        for(k = 7; k > 0; k--)
261
2422560
            ptr0[k] = ptr0[k - 1];
262
346080
        ptr0[0] = out[i];
263
    }
264
265
51912
    for(i = 0; i < 8; i++)
266
46144
        t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15;
267
268
5768
    ptr0 = dec->tmp2;
269
351848
    for(i = 0; i < 60; i++){
270
346080
        int sum = 0;
271
3114720
        for(k = 0; k < 8; k++)
272
2768640
            sum += ptr0[k] * t[k];
273
2768640
        for(k = 7; k > 0; k--)
274
2422560
            ptr0[k] = ptr0[k - 1];
275
346080
        ptr0[0] = out[i];
276
346080
        out[i] += (- sum) >> 12;
277
    }
278
279
51912
    for(i = 0; i < 8; i++)
280
46144
        t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15;
281
282
5768
    ptr0 = dec->tmp3;
283
351848
    for(i = 0; i < 60; i++){
284
346080
        int sum = out[i] * (1 << 12);
285
3114720
        for(k = 0; k < 8; k++)
286
2768640
            sum += ptr0[k] * t[k];
287
2768640
        for(k = 7; k > 0; k--)
288
2422560
            ptr0[k] = ptr0[k - 1];
289
346080
        ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
290
291
346080
        sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
292
346080
        sum = sum - (sum >> 3);
293
346080
        out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
294
    }
295
5768
}
296
297
1442
static void truespeech_save_prevvec(TSContext *c)
298
{
299
    int i;
300
301
12978
    for(i = 0; i < 8; i++)
302
11536
        c->prevfilt[i] = c->cvector[i];
303
1442
}
304
305
12
static int truespeech_decode_frame(AVCodecContext *avctx, void *data,
306
                                   int *got_frame_ptr, AVPacket *avpkt)
307
{
308
12
    AVFrame *frame     = data;
309
12
    const uint8_t *buf = avpkt->data;
310
12
    int buf_size = avpkt->size;
311
12
    TSContext *c = avctx->priv_data;
312
313
    int i, j;
314
    int16_t *samples;
315
    int iterations, ret;
316
317
12
    iterations = buf_size / 32;
318
319
12
    if (!iterations) {
320
        av_log(avctx, AV_LOG_ERROR,
321
               "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size);
322
        return -1;
323
    }
324
325
    /* get output buffer */
326
12
    frame->nb_samples = iterations * 240;
327
12
    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
328
        return ret;
329
12
    samples = (int16_t *)frame->data[0];
330
331
12
    memset(samples, 0, iterations * 240 * sizeof(*samples));
332
333
1454
    for(j = 0; j < iterations; j++) {
334
1442
        truespeech_read_frame(c, buf);
335
1442
        buf += 32;
336
337
1442
        truespeech_correlate_filter(c);
338
1442
        truespeech_filters_merge(c);
339
340
7210
        for(i = 0; i < 4; i++) {
341
5768
            truespeech_apply_twopoint_filter(c, i);
342
5768
            truespeech_place_pulses  (c, samples, i);
343
5768
            truespeech_update_filters(c, samples, i);
344
5768
            truespeech_synth         (c, samples, i);
345
5768
            samples += 60;
346
        }
347
348
1442
        truespeech_save_prevvec(c);
349
    }
350
351
12
    *got_frame_ptr = 1;
352
353
12
    return buf_size;
354
}
355
356
AVCodec ff_truespeech_decoder = {
357
    .name           = "truespeech",
358
    .long_name      = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"),
359
    .type           = AVMEDIA_TYPE_AUDIO,
360
    .id             = AV_CODEC_ID_TRUESPEECH,
361
    .priv_data_size = sizeof(TSContext),
362
    .init           = truespeech_decode_init,
363
    .decode         = truespeech_decode_frame,
364
    .capabilities   = AV_CODEC_CAP_DR1,
365
};