Browse code

Add native GSM 06.10 audio decoder.

Originally committed as revision 24158 to svn://svn.ffmpeg.org/ffmpeg/trunk

Reimar Döffinger authored on 2010/07/10 16:55:06
Showing 6 changed files
... ...
@@ -19,6 +19,7 @@ version <next>:
19 19
 - RTP depacketization of SVQ3
20 20
 - -strict inofficial replaced by -strict unofficial
21 21
 - ffplay -exitonkeydown and -exitonmousedown options added
22
+- native GSM / GSM MS decoder
22 23
 
23 24
 
24 25
 
... ...
@@ -589,10 +589,10 @@ following image formats are supported:
589 589
 @item Enhanced AC-3          @tab     @tab  X
590 590
 @item FLAC (Free Lossless Audio Codec)  @tab  X  @tab  IX
591 591
 @item G.729                  @tab     @tab  X
592
-@item GSM                    @tab  E  @tab  E
593
-    @tab supported through external library libgsm
594
-@item GSM Microsoft variant  @tab  E  @tab  E
595
-    @tab supported through external library libgsm
592
+@item GSM                    @tab  E  @tab  X
593
+    @tab encoding supported through external library libgsm
594
+@item GSM Microsoft variant  @tab  E  @tab  X
595
+    @tab encoding supported through external library libgsm
596 596
 @item IMC (Intel Music Coder)  @tab     @tab  X
597 597
 @item MACE (Macintosh Audio Compression/Expansion) 3:1  @tab     @tab  X
598 598
 @item MACE (Macintosh Audio Compression/Expansion) 6:1  @tab     @tab  X
... ...
@@ -130,6 +130,8 @@ OBJS-$(CONFIG_FRAPS_DECODER)           += fraps.o huffman.o
130 130
 OBJS-$(CONFIG_FRWU_DECODER)            += frwu.o
131 131
 OBJS-$(CONFIG_GIF_DECODER)             += gifdec.o lzw.o
132 132
 OBJS-$(CONFIG_GIF_ENCODER)             += gif.o lzwenc.o
133
+OBJS-$(CONFIG_GSM_DECODER)             += gsmdec.o
134
+OBJS-$(CONFIG_GSM_MS_DECODER)          += gsmdec.o
133 135
 OBJS-$(CONFIG_H261_DECODER)            += h261dec.o h261.o \
134 136
                                           mpegvideo.o error_resilience.o
135 137
 OBJS-$(CONFIG_H261_ENCODER)            += h261enc.o h261.o             \
... ...
@@ -230,6 +230,8 @@ void avcodec_register_all(void)
230 230
     REGISTER_DECODER (DSICINAUDIO, dsicinaudio);
231 231
     REGISTER_DECODER (EAC3, eac3);
232 232
     REGISTER_ENCDEC  (FLAC, flac);
233
+    REGISTER_DECODER (GSM, gsm);
234
+    REGISTER_DECODER (GSM_MS, gsm_ms);
233 235
     REGISTER_DECODER (IMC, imc);
234 236
     REGISTER_DECODER (MACE3, mace3);
235 237
     REGISTER_DECODER (MACE6, mace6);
... ...
@@ -30,7 +30,7 @@
30 30
 #include "libavutil/avutil.h"
31 31
 
32 32
 #define LIBAVCODEC_VERSION_MAJOR 52
33
-#define LIBAVCODEC_VERSION_MINOR 80
33
+#define LIBAVCODEC_VERSION_MINOR 81
34 34
 #define LIBAVCODEC_VERSION_MICRO  0
35 35
 
36 36
 #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
37 37
new file mode 100644
... ...
@@ -0,0 +1,316 @@
0
+/*
1
+ * gsm 06.10 decoder
2
+ * Copyright (c) 2010 Reimar Döffinger <Reimar.Doeffinger@gmx.de>
3
+ *
4
+ * This file is part of FFmpeg.
5
+ *
6
+ * FFmpeg is free software; you can redistribute it and/or
7
+ * modify it under the terms of the GNU Lesser General Public
8
+ * License as published by the Free Software Foundation; either
9
+ * version 2.1 of the License, or (at your option) any later version.
10
+ *
11
+ * FFmpeg is distributed in the hope that it will be useful,
12
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
13
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14
+ * Lesser General Public License for more details.
15
+ *
16
+ * You should have received a copy of the GNU Lesser General Public
17
+ * License along with FFmpeg; if not, write to the Free Software
18
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
+ */
20
+
21
+/**
22
+ * @file
23
+ * GSM decoder
24
+ */
25
+
26
+#define ALT_BITSTREAM_READER_LE
27
+#include "avcodec.h"
28
+#include "get_bits.h"
29
+
30
+// input and output sizes in byte
31
+#define GSM_BLOCK_SIZE    33
32
+#define GSM_MS_BLOCK_SIZE 65
33
+#define GSM_FRAME_SIZE   160
34
+
35
+typedef struct {
36
+    int16_t ref_buf[280];
37
+    int v[9];
38
+    int lar[2][8];
39
+    int lar_idx;
40
+    int msr;
41
+} GSMContext;
42
+
43
+static av_cold int gsm_init(AVCodecContext *avctx)
44
+{
45
+    avctx->channels = 1;
46
+    if (!avctx->sample_rate)
47
+        avctx->sample_rate = 8000;
48
+    avctx->sample_fmt = SAMPLE_FMT_S16;
49
+
50
+    switch (avctx->codec_id) {
51
+    case CODEC_ID_GSM:
52
+        avctx->frame_size  = GSM_FRAME_SIZE;
53
+        avctx->block_align = GSM_BLOCK_SIZE;
54
+        break;
55
+    case CODEC_ID_GSM_MS:
56
+        avctx->frame_size  = 2 * GSM_FRAME_SIZE;
57
+        avctx->block_align = GSM_MS_BLOCK_SIZE;
58
+    }
59
+
60
+    return 0;
61
+}
62
+
63
+static const int16_t dequant_tab[64][8] = {
64
+    {   -28,    -20,    -12,     -4,      4,     12,     20,     28},
65
+    {   -56,    -40,    -24,     -8,      8,     24,     40,     56},
66
+    {   -84,    -60,    -36,    -12,     12,     36,     60,     84},
67
+    {  -112,    -80,    -48,    -16,     16,     48,     80,    112},
68
+    {  -140,   -100,    -60,    -20,     20,     60,    100,    140},
69
+    {  -168,   -120,    -72,    -24,     24,     72,    120,    168},
70
+    {  -196,   -140,    -84,    -28,     28,     84,    140,    196},
71
+    {  -224,   -160,    -96,    -32,     32,     96,    160,    224},
72
+    {  -252,   -180,   -108,    -36,     36,    108,    180,    252},
73
+    {  -280,   -200,   -120,    -40,     40,    120,    200,    280},
74
+    {  -308,   -220,   -132,    -44,     44,    132,    220,    308},
75
+    {  -336,   -240,   -144,    -48,     48,    144,    240,    336},
76
+    {  -364,   -260,   -156,    -52,     52,    156,    260,    364},
77
+    {  -392,   -280,   -168,    -56,     56,    168,    280,    392},
78
+    {  -420,   -300,   -180,    -60,     60,    180,    300,    420},
79
+    {  -448,   -320,   -192,    -64,     64,    192,    320,    448},
80
+    {  -504,   -360,   -216,    -72,     72,    216,    360,    504},
81
+    {  -560,   -400,   -240,    -80,     80,    240,    400,    560},
82
+    {  -616,   -440,   -264,    -88,     88,    264,    440,    616},
83
+    {  -672,   -480,   -288,    -96,     96,    288,    480,    672},
84
+    {  -728,   -520,   -312,   -104,    104,    312,    520,    728},
85
+    {  -784,   -560,   -336,   -112,    112,    336,    560,    784},
86
+    {  -840,   -600,   -360,   -120,    120,    360,    600,    840},
87
+    {  -896,   -640,   -384,   -128,    128,    384,    640,    896},
88
+    { -1008,   -720,   -432,   -144,    144,    432,    720,   1008},
89
+    { -1120,   -800,   -480,   -160,    160,    480,    800,   1120},
90
+    { -1232,   -880,   -528,   -176,    176,    528,    880,   1232},
91
+    { -1344,   -960,   -576,   -192,    192,    576,    960,   1344},
92
+    { -1456,  -1040,   -624,   -208,    208,    624,   1040,   1456},
93
+    { -1568,  -1120,   -672,   -224,    224,    672,   1120,   1568},
94
+    { -1680,  -1200,   -720,   -240,    240,    720,   1200,   1680},
95
+    { -1792,  -1280,   -768,   -256,    256,    768,   1280,   1792},
96
+    { -2016,  -1440,   -864,   -288,    288,    864,   1440,   2016},
97
+    { -2240,  -1600,   -960,   -320,    320,    960,   1600,   2240},
98
+    { -2464,  -1760,  -1056,   -352,    352,   1056,   1760,   2464},
99
+    { -2688,  -1920,  -1152,   -384,    384,   1152,   1920,   2688},
100
+    { -2912,  -2080,  -1248,   -416,    416,   1248,   2080,   2912},
101
+    { -3136,  -2240,  -1344,   -448,    448,   1344,   2240,   3136},
102
+    { -3360,  -2400,  -1440,   -480,    480,   1440,   2400,   3360},
103
+    { -3584,  -2560,  -1536,   -512,    512,   1536,   2560,   3584},
104
+    { -4032,  -2880,  -1728,   -576,    576,   1728,   2880,   4032},
105
+    { -4480,  -3200,  -1920,   -640,    640,   1920,   3200,   4480},
106
+    { -4928,  -3520,  -2112,   -704,    704,   2112,   3520,   4928},
107
+    { -5376,  -3840,  -2304,   -768,    768,   2304,   3840,   5376},
108
+    { -5824,  -4160,  -2496,   -832,    832,   2496,   4160,   5824},
109
+    { -6272,  -4480,  -2688,   -896,    896,   2688,   4480,   6272},
110
+    { -6720,  -4800,  -2880,   -960,    960,   2880,   4800,   6720},
111
+    { -7168,  -5120,  -3072,  -1024,   1024,   3072,   5120,   7168},
112
+    { -8063,  -5759,  -3456,  -1152,   1152,   3456,   5760,   8064},
113
+    { -8959,  -6399,  -3840,  -1280,   1280,   3840,   6400,   8960},
114
+    { -9855,  -7039,  -4224,  -1408,   1408,   4224,   7040,   9856},
115
+    {-10751,  -7679,  -4608,  -1536,   1536,   4608,   7680,  10752},
116
+    {-11647,  -8319,  -4992,  -1664,   1664,   4992,   8320,  11648},
117
+    {-12543,  -8959,  -5376,  -1792,   1792,   5376,   8960,  12544},
118
+    {-13439,  -9599,  -5760,  -1920,   1920,   5760,   9600,  13440},
119
+    {-14335, -10239,  -6144,  -2048,   2048,   6144,  10240,  14336},
120
+    {-16127, -11519,  -6912,  -2304,   2304,   6912,  11519,  16127},
121
+    {-17919, -12799,  -7680,  -2560,   2560,   7680,  12799,  17919},
122
+    {-19711, -14079,  -8448,  -2816,   2816,   8448,  14079,  19711},
123
+    {-21503, -15359,  -9216,  -3072,   3072,   9216,  15359,  21503},
124
+    {-23295, -16639,  -9984,  -3328,   3328,   9984,  16639,  23295},
125
+    {-25087, -17919, -10752,  -3584,   3584,  10752,  17919,  25087},
126
+    {-26879, -19199, -11520,  -3840,   3840,  11520,  19199,  26879},
127
+    {-28671, -20479, -12288,  -4096,   4096,  12288,  20479,  28671}
128
+};
129
+
130
+static void apcm_dequant_add(GetBitContext *gb, int16_t *dst)
131
+{
132
+    int i;
133
+    int maxidx = get_bits(gb, 6);
134
+    const int16_t *tab = dequant_tab[maxidx];
135
+    for (i = 0; i < 13; i++)
136
+        dst[3*i] += tab[get_bits(gb, 3)];
137
+}
138
+
139
+static inline int gsm_mult(int a, int b)
140
+{
141
+    return (a * b + (1 << 14)) >> 15;
142
+}
143
+
144
+static const uint16_t long_term_gain_tab[4] = {
145
+    3277, 11469, 21299, 32767
146
+};
147
+
148
+static void long_term_synth(int16_t *dst, int lag, int gain_idx)
149
+{
150
+    int i;
151
+    const int16_t *src = dst - lag;
152
+    uint16_t gain = long_term_gain_tab[gain_idx];
153
+    for (i = 0; i < 40; i++)
154
+        dst[i] = gsm_mult(gain, src[i]);
155
+}
156
+
157
+static inline int decode_log_area(int coded, int factor, int offset)
158
+{
159
+    coded <<= 10;
160
+    coded -= offset;
161
+    return gsm_mult(coded, factor) << 1;
162
+}
163
+
164
+static av_noinline int get_rrp(int filtered)
165
+{
166
+    int abs = FFABS(filtered);
167
+    if      (abs < 11059) abs <<= 1;
168
+    else if (abs < 20070) abs += 11059;
169
+    else                  abs = (abs >> 2) + 26112;
170
+    return filtered < 0 ? -abs : abs;
171
+}
172
+
173
+static int filter_value(int in, int rrp[8], int v[9])
174
+{
175
+    int i;
176
+    for (i = 7; i >= 0; i--) {
177
+        in -= gsm_mult(rrp[i], v[i]);
178
+        v[i + 1] = v[i] + gsm_mult(rrp[i], in);
179
+    }
180
+    v[0] = in;
181
+    return in;
182
+}
183
+
184
+static void short_term_synth(GSMContext *ctx, int16_t *dst, const int16_t *src)
185
+{
186
+    int i;
187
+    int rrp[8];
188
+    int *lar = ctx->lar[ctx->lar_idx];
189
+    int *lar_prev = ctx->lar[ctx->lar_idx ^ 1];
190
+    for (i = 0; i < 8; i++)
191
+        rrp[i] = get_rrp((lar_prev[i] >> 2) + (lar_prev[i] >> 1) + (lar[i] >> 2));
192
+    for (i = 0; i < 13; i++)
193
+        dst[i] = filter_value(src[i], rrp, ctx->v);
194
+
195
+    for (i = 0; i < 8; i++)
196
+        rrp[i] = get_rrp((lar_prev[i] >> 1) + (lar     [i] >> 1));
197
+    for (i = 13; i < 27; i++)
198
+        dst[i] = filter_value(src[i], rrp, ctx->v);
199
+
200
+    for (i = 0; i < 8; i++)
201
+        rrp[i] = get_rrp((lar_prev[i] >> 2) + (lar     [i] >> 1) + (lar[i] >> 2));
202
+    for (i = 27; i < 40; i++)
203
+        dst[i] = filter_value(src[i], rrp, ctx->v);
204
+
205
+    for (i = 0; i < 8; i++)
206
+        rrp[i] = get_rrp(lar[i]);
207
+    for (i = 40; i < 160; i++)
208
+        dst[i] = filter_value(src[i], rrp, ctx->v);
209
+
210
+    ctx->lar_idx ^= 1;
211
+}
212
+
213
+static int postprocess(int16_t *data, int msr)
214
+{
215
+    int i;
216
+    for (i = 0; i < 160; i++) {
217
+        msr = av_clip_int16(data[i] + gsm_mult(msr, 28180));
218
+        data[i] = av_clip_int16(msr << 1) & ~7;
219
+    }
220
+    return msr;
221
+}
222
+
223
+static int gsm_decode_block(AVCodecContext *avctx, int16_t *samples,
224
+                            GetBitContext *gb)
225
+{
226
+    GSMContext *ctx = avctx->priv_data;
227
+    int i;
228
+    int16_t *ref_dst = ctx->ref_buf + 120;
229
+    int *lar = ctx->lar[ctx->lar_idx];
230
+    lar[0] = decode_log_area(get_bits(gb, 6), 13107,  1 << 15);
231
+    lar[1] = decode_log_area(get_bits(gb, 6), 13107,  1 << 15);
232
+    lar[2] = decode_log_area(get_bits(gb, 5), 13107, (1 << 14) + 2048*2);
233
+    lar[3] = decode_log_area(get_bits(gb, 5), 13107, (1 << 14) - 2560*2);
234
+    lar[4] = decode_log_area(get_bits(gb, 4), 19223, (1 << 13) +   94*2);
235
+    lar[5] = decode_log_area(get_bits(gb, 4), 17476, (1 << 13) - 1792*2);
236
+    lar[6] = decode_log_area(get_bits(gb, 3), 31454, (1 << 12) -  341*2);
237
+    lar[7] = decode_log_area(get_bits(gb, 3), 29708, (1 << 12) - 1144*2);
238
+
239
+    for (i = 0; i < 4; i++) {
240
+        int lag      = get_bits(gb, 7);
241
+        int gain_idx = get_bits(gb, 2);
242
+        int offset   = get_bits(gb, 2);
243
+        lag = av_clip(lag, 40, 120);
244
+        long_term_synth(ref_dst, lag, gain_idx);
245
+        apcm_dequant_add(gb, ref_dst + offset);
246
+        ref_dst += 40;
247
+    }
248
+    memcpy(ctx->ref_buf, ctx->ref_buf + 160, 120 * sizeof(*ctx->ref_buf));
249
+    short_term_synth(ctx, samples, ctx->ref_buf + 120);
250
+    // for optimal speed this could be merged with short_term_synth,
251
+    // not done yet because it is a bit ugly
252
+    ctx->msr = postprocess(samples, ctx->msr);
253
+    return 0;
254
+}
255
+
256
+static int gsm_decode_frame(AVCodecContext *avctx, void *data,
257
+                            int *data_size, AVPacket *avpkt)
258
+{
259
+    int res;
260
+    GetBitContext gb;
261
+    const uint8_t *buf = avpkt->data;
262
+    int buf_size = avpkt->size;
263
+    int16_t *samples = data;
264
+    int frame_bytes = 2 * avctx->frame_size;
265
+
266
+    if (*data_size < frame_bytes)
267
+        return -1;
268
+    *data_size = 0;
269
+    if(buf_size < avctx->block_align)
270
+        return AVERROR_INVALIDDATA;
271
+    init_get_bits(&gb, buf, buf_size * 8);
272
+
273
+    switch (avctx->codec_id) {
274
+    case CODEC_ID_GSM:
275
+        if (get_bits(&gb, 4) != 0xd)
276
+            av_log(avctx, AV_LOG_WARNING, "Missing GSM magic!\n");
277
+        res = gsm_decode_block(avctx, samples, &gb);
278
+        if (res < 0)
279
+            return res;
280
+        break;
281
+    case CODEC_ID_GSM_MS:
282
+        res = gsm_decode_block(avctx, samples, &gb);
283
+        if (res < 0)
284
+            return res;
285
+        res = gsm_decode_block(avctx, samples + GSM_FRAME_SIZE, &gb);
286
+        if (res < 0)
287
+            return res;
288
+    }
289
+    *data_size = frame_bytes;
290
+    return avctx->block_align;
291
+}
292
+
293
+AVCodec gsm_decoder = {
294
+    "gsm",
295
+    AVMEDIA_TYPE_AUDIO,
296
+    CODEC_ID_GSM,
297
+    sizeof(GSMContext),
298
+    gsm_init,
299
+    NULL,
300
+    NULL,
301
+    gsm_decode_frame,
302
+    .long_name = NULL_IF_CONFIG_SMALL("GSM"),
303
+};
304
+
305
+AVCodec gsm_ms_decoder = {
306
+    "gsm_ms",
307
+    AVMEDIA_TYPE_AUDIO,
308
+    CODEC_ID_GSM_MS,
309
+    sizeof(GSMContext),
310
+    gsm_init,
311
+    NULL,
312
+    NULL,
313
+    gsm_decode_frame,
314
+    .long_name = NULL_IF_CONFIG_SMALL("GSM Microsoft variant"),
315
+};