2 * Windows Media Audio Voice decoder.
3 * Copyright (c) 2009 Ronald S. Bultje
5 * This file is part of FFmpeg.
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.
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.
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
24 * @brief Windows Media Audio Voice compatible decoder
25 * @author Ronald S. Bultje <rsbultje@gmail.com>
30 #include "libavutil/channel_layout.h"
31 #include "libavutil/float_dsp.h"
32 #include "libavutil/mem_internal.h"
33 #include "libavutil/thread.h"
38 #include "wmavoice_data.h"
39 #include "celp_filters.h"
40 #include "acelp_vectors.h"
41 #include "acelp_filters.h"
47 #define MAX_BLOCKS 8 ///< maximum number of blocks per frame
48 #define MAX_LSPS 16 ///< maximum filter order
49 #define MAX_LSPS_ALIGN16 16 ///< same as #MAX_LSPS; needs to be multiple
50 ///< of 16 for ASM input buffer alignment
51 #define MAX_FRAMES 3 ///< maximum number of frames per superframe
52 #define MAX_FRAMESIZE 160 ///< maximum number of samples per frame
53 #define MAX_SIGNAL_HISTORY 416 ///< maximum excitation signal history
54 #define MAX_SFRAMESIZE (MAX_FRAMESIZE * MAX_FRAMES)
55 ///< maximum number of samples per superframe
56 #define SFRAME_CACHE_MAXSIZE 256 ///< maximum cache size for frame data that
57 ///< was split over two packets
58 #define VLC_NBITS 6 ///< number of bits to read per VLC iteration
61 * Frame type VLC coding.
63 static VLC frame_type_vlc;
66 * Adaptive codebook types.
69 ACB_TYPE_NONE = 0, ///< no adaptive codebook (only hardcoded fixed)
70 ACB_TYPE_ASYMMETRIC = 1, ///< adaptive codebook with per-frame pitch, which
71 ///< we interpolate to get a per-sample pitch.
72 ///< Signal is generated using an asymmetric sinc
74 ///< @note see #wmavoice_ipol1_coeffs
75 ACB_TYPE_HAMMING = 2 ///< Per-block pitch with signal generation using
76 ///< a Hamming sinc window function
77 ///< @note see #wmavoice_ipol2_coeffs
81 * Fixed codebook types.
84 FCB_TYPE_SILENCE = 0, ///< comfort noise during silence
85 ///< generated from a hardcoded (fixed) codebook
86 ///< with per-frame (low) gain values
87 FCB_TYPE_HARDCODED = 1, ///< hardcoded (fixed) codebook with per-block
89 FCB_TYPE_AW_PULSES = 2, ///< Pitch-adaptive window (AW) pulse signals,
90 ///< used in particular for low-bitrate streams
91 FCB_TYPE_EXC_PULSES = 3, ///< Innovation (fixed) codebook pulse sets in
92 ///< combinations of either single pulses or
97 * Description of frame types.
99 static const struct frame_type_desc {
100 uint8_t n_blocks; ///< amount of blocks per frame (each block
101 ///< (contains 160/#n_blocks samples)
102 uint8_t log_n_blocks; ///< log2(#n_blocks)
103 uint8_t acb_type; ///< Adaptive codebook type (ACB_TYPE_*)
104 uint8_t fcb_type; ///< Fixed codebook type (FCB_TYPE_*)
105 uint8_t dbl_pulses; ///< how many pulse vectors have pulse pairs
106 ///< (rather than just one single pulse)
107 ///< only if #fcb_type == #FCB_TYPE_EXC_PULSES
108 } frame_descs[17] = {
109 { 1, 0, ACB_TYPE_NONE, FCB_TYPE_SILENCE, 0 },
110 { 2, 1, ACB_TYPE_NONE, FCB_TYPE_HARDCODED, 0 },
111 { 2, 1, ACB_TYPE_ASYMMETRIC, FCB_TYPE_AW_PULSES, 0 },
112 { 2, 1, ACB_TYPE_ASYMMETRIC, FCB_TYPE_EXC_PULSES, 2 },
113 { 2, 1, ACB_TYPE_ASYMMETRIC, FCB_TYPE_EXC_PULSES, 5 },
114 { 4, 2, ACB_TYPE_ASYMMETRIC, FCB_TYPE_EXC_PULSES, 0 },
115 { 4, 2, ACB_TYPE_ASYMMETRIC, FCB_TYPE_EXC_PULSES, 2 },
116 { 4, 2, ACB_TYPE_ASYMMETRIC, FCB_TYPE_EXC_PULSES, 5 },
117 { 2, 1, ACB_TYPE_HAMMING, FCB_TYPE_EXC_PULSES, 0 },
118 { 2, 1, ACB_TYPE_HAMMING, FCB_TYPE_EXC_PULSES, 2 },
119 { 2, 1, ACB_TYPE_HAMMING, FCB_TYPE_EXC_PULSES, 5 },
120 { 4, 2, ACB_TYPE_HAMMING, FCB_TYPE_EXC_PULSES, 0 },
121 { 4, 2, ACB_TYPE_HAMMING, FCB_TYPE_EXC_PULSES, 2 },
122 { 4, 2, ACB_TYPE_HAMMING, FCB_TYPE_EXC_PULSES, 5 },
123 { 8, 3, ACB_TYPE_HAMMING, FCB_TYPE_EXC_PULSES, 0 },
124 { 8, 3, ACB_TYPE_HAMMING, FCB_TYPE_EXC_PULSES, 2 },
125 { 8, 3, ACB_TYPE_HAMMING, FCB_TYPE_EXC_PULSES, 5 }
129 * WMA Voice decoding context.
131 typedef struct WMAVoiceContext {
133 * @name Global values specified in the stream header / extradata or used all over.
136 GetBitContext gb; ///< packet bitreader. During decoder init,
137 ///< it contains the extradata from the
138 ///< demuxer. During decoding, it contains
140 int8_t vbm_tree[25]; ///< converts VLC codes to frame type
142 int spillover_bitsize; ///< number of bits used to specify
143 ///< #spillover_nbits in the packet header
144 ///< = ceil(log2(ctx->block_align << 3))
145 int history_nsamples; ///< number of samples in history for signal
146 ///< prediction (through ACB)
148 /* postfilter specific values */
149 int do_apf; ///< whether to apply the averaged
150 ///< projection filter (APF)
151 int denoise_strength; ///< strength of denoising in Wiener filter
153 int denoise_tilt_corr; ///< Whether to apply tilt correction to the
154 ///< Wiener filter coefficients (postfilter)
155 int dc_level; ///< Predicted amount of DC noise, based
156 ///< on which a DC removal filter is used
158 int lsps; ///< number of LSPs per frame [10 or 16]
159 int lsp_q_mode; ///< defines quantizer defaults [0, 1]
160 int lsp_def_mode; ///< defines different sets of LSP defaults
163 int min_pitch_val; ///< base value for pitch parsing code
164 int max_pitch_val; ///< max value + 1 for pitch parsing
165 int pitch_nbits; ///< number of bits used to specify the
166 ///< pitch value in the frame header
167 int block_pitch_nbits; ///< number of bits used to specify the
168 ///< first block's pitch value
169 int block_pitch_range; ///< range of the block pitch
170 int block_delta_pitch_nbits; ///< number of bits used to specify the
171 ///< delta pitch between this and the last
172 ///< block's pitch value, used in all but
174 int block_delta_pitch_hrange; ///< 1/2 range of the delta (full range is
175 ///< from -this to +this-1)
176 uint16_t block_conv_table[4]; ///< boundaries for block pitch unit/scale
182 * @name Packet values specified in the packet header or related to a packet.
184 * A packet is considered to be a single unit of data provided to this
185 * decoder by the demuxer.
188 int spillover_nbits; ///< number of bits of the previous packet's
189 ///< last superframe preceding this
190 ///< packet's first full superframe (useful
191 ///< for re-synchronization also)
192 int has_residual_lsps; ///< if set, superframes contain one set of
193 ///< LSPs that cover all frames, encoded as
194 ///< independent and residual LSPs; if not
195 ///< set, each frame contains its own, fully
196 ///< independent, LSPs
197 int skip_bits_next; ///< number of bits to skip at the next call
198 ///< to #wmavoice_decode_packet() (since
199 ///< they're part of the previous superframe)
201 uint8_t sframe_cache[SFRAME_CACHE_MAXSIZE + AV_INPUT_BUFFER_PADDING_SIZE];
202 ///< cache for superframe data split over
203 ///< multiple packets
204 int sframe_cache_size; ///< set to >0 if we have data from an
205 ///< (incomplete) superframe from a previous
206 ///< packet that spilled over in the current
207 ///< packet; specifies the amount of bits in
209 PutBitContext pb; ///< bitstream writer for #sframe_cache
214 * @name Frame and superframe values
215 * Superframe and frame data - these can change from frame to frame,
216 * although some of them do in that case serve as a cache / history for
217 * the next frame or superframe.
220 double prev_lsps[MAX_LSPS]; ///< LSPs of the last frame of the previous
222 int last_pitch_val; ///< pitch value of the previous frame
223 int last_acb_type; ///< frame type [0-2] of the previous frame
224 int pitch_diff_sh16; ///< ((cur_pitch_val - #last_pitch_val)
225 ///< << 16) / #MAX_FRAMESIZE
226 float silence_gain; ///< set for use in blocks if #ACB_TYPE_NONE
228 int aw_idx_is_ext; ///< whether the AW index was encoded in
229 ///< 8 bits (instead of 6)
230 int aw_pulse_range; ///< the range over which #aw_pulse_set1()
231 ///< can apply the pulse, relative to the
232 ///< value in aw_first_pulse_off. The exact
233 ///< position of the first AW-pulse is within
234 ///< [pulse_off, pulse_off + this], and
235 ///< depends on bitstream values; [16 or 24]
236 int aw_n_pulses[2]; ///< number of AW-pulses in each block; note
237 ///< that this number can be negative (in
238 ///< which case it basically means "zero")
239 int aw_first_pulse_off[2]; ///< index of first sample to which to
240 ///< apply AW-pulses, or -0xff if unset
241 int aw_next_pulse_off_cache; ///< the position (relative to start of the
242 ///< second block) at which pulses should
243 ///< start to be positioned, serves as a
244 ///< cache for pitch-adaptive window pulses
247 int frame_cntr; ///< current frame index [0 - 0xFFFE]; is
248 ///< only used for comfort noise in #pRNG()
249 int nb_superframes; ///< number of superframes in current packet
250 float gain_pred_err[6]; ///< cache for gain prediction
251 float excitation_history[MAX_SIGNAL_HISTORY];
252 ///< cache of the signal of previous
253 ///< superframes, used as a history for
254 ///< signal generation
255 float synth_history[MAX_LSPS]; ///< see #excitation_history
259 * @name Postfilter values
261 * Variables used for postfilter implementation, mostly history for
262 * smoothing and so on, and context variables for FFT/iFFT.
265 RDFTContext rdft, irdft; ///< contexts for FFT-calculation in the
266 ///< postfilter (for denoise filter)
267 DCTContext dct, dst; ///< contexts for phase shift (in Hilbert
268 ///< transform, part of postfilter)
269 float sin[511], cos[511]; ///< 8-bit cosine/sine windows over [-pi,pi]
271 float postfilter_agc; ///< gain control memory, used in
272 ///< #adaptive_gain_control()
273 float dcf_mem[2]; ///< DC filter history
274 float zero_exc_pf[MAX_SIGNAL_HISTORY + MAX_SFRAMESIZE];
275 ///< zero filter output (i.e. excitation)
277 float denoise_filter_cache[MAX_FRAMESIZE];
278 int denoise_filter_cache_size; ///< samples in #denoise_filter_cache
279 DECLARE_ALIGNED(32, float, tilted_lpcs_pf)[0x80];
280 ///< aligned buffer for LPC tilting
281 DECLARE_ALIGNED(32, float, denoise_coeffs_pf)[0x80];
282 ///< aligned buffer for denoise coefficients
283 DECLARE_ALIGNED(32, float, synth_filter_out_buf)[0x80 + MAX_LSPS_ALIGN16];
284 ///< aligned buffer for postfilter speech
292 * Set up the variable bit mode (VBM) tree from container extradata.
293 * @param gb bit I/O context.
294 * The bit context (s->gb) should be loaded with byte 23-46 of the
295 * container extradata (i.e. the ones containing the VBM tree).
296 * @param vbm_tree pointer to array to which the decoded VBM tree will be
298 * @return 0 on success, <0 on error.
300 static av_cold int decode_vbmtree(GetBitContext *gb, int8_t vbm_tree[25])
302 int cntr[8] = { 0 }, n, res;
304 memset(vbm_tree, 0xff, sizeof(vbm_tree[0]) * 25);
305 for (n = 0; n < 17; n++) {
306 res = get_bits(gb, 3);
307 if (cntr[res] > 3) // should be >= 3 + (res == 7))
309 vbm_tree[res * 3 + cntr[res]++] = n;
314 static av_cold void wmavoice_init_static_data(void)
316 static const uint8_t bits[] = {
319 10, 10, 10, 12, 12, 12,
322 static const uint16_t codes[] = {
323 0x0000, 0x0001, 0x0002, // 00/01/10
324 0x000c, 0x000d, 0x000e, // 11+00/01/10
325 0x003c, 0x003d, 0x003e, // 1111+00/01/10
326 0x00fc, 0x00fd, 0x00fe, // 111111+00/01/10
327 0x03fc, 0x03fd, 0x03fe, // 11111111+00/01/10
328 0x0ffc, 0x0ffd, 0x0ffe, // 1111111111+00/01/10
329 0x3ffc, 0x3ffd, 0x3ffe, 0x3fff // 111111111111+xx
332 INIT_VLC_STATIC(&frame_type_vlc, VLC_NBITS, sizeof(bits),
333 bits, 1, 1, codes, 2, 2, 132);
336 static av_cold void wmavoice_flush(AVCodecContext *ctx)
338 WMAVoiceContext *s = ctx->priv_data;
341 s->postfilter_agc = 0;
342 s->sframe_cache_size = 0;
343 s->skip_bits_next = 0;
344 for (n = 0; n < s->lsps; n++)
345 s->prev_lsps[n] = M_PI * (n + 1.0) / (s->lsps + 1.0);
346 memset(s->excitation_history, 0,
347 sizeof(*s->excitation_history) * MAX_SIGNAL_HISTORY);
348 memset(s->synth_history, 0,
349 sizeof(*s->synth_history) * MAX_LSPS);
350 memset(s->gain_pred_err, 0,
351 sizeof(s->gain_pred_err));
354 memset(&s->synth_filter_out_buf[MAX_LSPS_ALIGN16 - s->lsps], 0,
355 sizeof(*s->synth_filter_out_buf) * s->lsps);
356 memset(s->dcf_mem, 0,
357 sizeof(*s->dcf_mem) * 2);
358 memset(s->zero_exc_pf, 0,
359 sizeof(*s->zero_exc_pf) * s->history_nsamples);
360 memset(s->denoise_filter_cache, 0, sizeof(s->denoise_filter_cache));
365 * Set up decoder with parameters from demuxer (extradata etc.).
367 static av_cold int wmavoice_decode_init(AVCodecContext *ctx)
369 static AVOnce init_static_once = AV_ONCE_INIT;
370 int n, flags, pitch_range, lsp16_flag;
371 WMAVoiceContext *s = ctx->priv_data;
373 ff_thread_once(&init_static_once, wmavoice_init_static_data);
377 * - byte 0-18: WMAPro-in-WMAVoice extradata (see wmaprodec.c),
378 * - byte 19-22: flags field (annoyingly in LE; see below for known
380 * - byte 23-46: variable bitmode tree (really just 17 * 3 bits,
383 if (ctx->extradata_size != 46) {
384 av_log(ctx, AV_LOG_ERROR,
385 "Invalid extradata size %d (should be 46)\n",
386 ctx->extradata_size);
387 return AVERROR_INVALIDDATA;
389 if (ctx->block_align <= 0 || ctx->block_align > (1<<22)) {
390 av_log(ctx, AV_LOG_ERROR, "Invalid block alignment %d.\n", ctx->block_align);
391 return AVERROR_INVALIDDATA;
394 flags = AV_RL32(ctx->extradata + 18);
395 s->spillover_bitsize = 3 + av_ceil_log2(ctx->block_align);
396 s->do_apf = flags & 0x1;
398 ff_rdft_init(&s->rdft, 7, DFT_R2C);
399 ff_rdft_init(&s->irdft, 7, IDFT_C2R);
400 ff_dct_init(&s->dct, 6, DCT_I);
401 ff_dct_init(&s->dst, 6, DST_I);
403 ff_sine_window_init(s->cos, 256);
404 memcpy(&s->sin[255], s->cos, 256 * sizeof(s->cos[0]));
405 for (n = 0; n < 255; n++) {
406 s->sin[n] = -s->sin[510 - n];
407 s->cos[510 - n] = s->cos[n];
410 s->denoise_strength = (flags >> 2) & 0xF;
411 if (s->denoise_strength >= 12) {
412 av_log(ctx, AV_LOG_ERROR,
413 "Invalid denoise filter strength %d (max=11)\n",
414 s->denoise_strength);
415 return AVERROR_INVALIDDATA;
417 s->denoise_tilt_corr = !!(flags & 0x40);
418 s->dc_level = (flags >> 7) & 0xF;
419 s->lsp_q_mode = !!(flags & 0x2000);
420 s->lsp_def_mode = !!(flags & 0x4000);
421 lsp16_flag = flags & 0x1000;
427 for (n = 0; n < s->lsps; n++)
428 s->prev_lsps[n] = M_PI * (n + 1.0) / (s->lsps + 1.0);
430 init_get_bits(&s->gb, ctx->extradata + 22, (ctx->extradata_size - 22) << 3);
431 if (decode_vbmtree(&s->gb, s->vbm_tree) < 0) {
432 av_log(ctx, AV_LOG_ERROR, "Invalid VBM tree; broken extradata?\n");
433 return AVERROR_INVALIDDATA;
436 if (ctx->sample_rate >= INT_MAX / (256 * 37))
437 return AVERROR_INVALIDDATA;
439 s->min_pitch_val = ((ctx->sample_rate << 8) / 400 + 50) >> 8;
440 s->max_pitch_val = ((ctx->sample_rate << 8) * 37 / 2000 + 50) >> 8;
441 pitch_range = s->max_pitch_val - s->min_pitch_val;
442 if (pitch_range <= 0) {
443 av_log(ctx, AV_LOG_ERROR, "Invalid pitch range; broken extradata?\n");
444 return AVERROR_INVALIDDATA;
446 s->pitch_nbits = av_ceil_log2(pitch_range);
447 s->last_pitch_val = 40;
448 s->last_acb_type = ACB_TYPE_NONE;
449 s->history_nsamples = s->max_pitch_val + 8;
451 if (s->min_pitch_val < 1 || s->history_nsamples > MAX_SIGNAL_HISTORY) {
452 int min_sr = ((((1 << 8) - 50) * 400) + 0xFF) >> 8,
453 max_sr = ((((MAX_SIGNAL_HISTORY - 8) << 8) + 205) * 2000 / 37) >> 8;
455 av_log(ctx, AV_LOG_ERROR,
456 "Unsupported samplerate %d (min=%d, max=%d)\n",
457 ctx->sample_rate, min_sr, max_sr); // 322-22097 Hz
459 return AVERROR(ENOSYS);
462 s->block_conv_table[0] = s->min_pitch_val;
463 s->block_conv_table[1] = (pitch_range * 25) >> 6;
464 s->block_conv_table[2] = (pitch_range * 44) >> 6;
465 s->block_conv_table[3] = s->max_pitch_val - 1;
466 s->block_delta_pitch_hrange = (pitch_range >> 3) & ~0xF;
467 if (s->block_delta_pitch_hrange <= 0) {
468 av_log(ctx, AV_LOG_ERROR, "Invalid delta pitch hrange; broken extradata?\n");
469 return AVERROR_INVALIDDATA;
471 s->block_delta_pitch_nbits = 1 + av_ceil_log2(s->block_delta_pitch_hrange);
472 s->block_pitch_range = s->block_conv_table[2] +
473 s->block_conv_table[3] + 1 +
474 2 * (s->block_conv_table[1] - 2 * s->min_pitch_val);
475 s->block_pitch_nbits = av_ceil_log2(s->block_pitch_range);
478 ctx->channel_layout = AV_CH_LAYOUT_MONO;
479 ctx->sample_fmt = AV_SAMPLE_FMT_FLT;
485 * @name Postfilter functions
486 * Postfilter functions (gain control, wiener denoise filter, DC filter,
487 * kalman smoothening, plus surrounding code to wrap it)
491 * Adaptive gain control (as used in postfilter).
493 * Identical to #ff_adaptive_gain_control() in acelp_vectors.c, except
494 * that the energy here is calculated using sum(abs(...)), whereas the
495 * other codecs (e.g. AMR-NB, SIPRO) use sqrt(dotproduct(...)).
497 * @param out output buffer for filtered samples
498 * @param in input buffer containing the samples as they are after the
499 * postfilter steps so far
500 * @param speech_synth input buffer containing speech synth before postfilter
501 * @param size input buffer size
502 * @param alpha exponential filter factor
503 * @param gain_mem pointer to filter memory (single float)
505 static void adaptive_gain_control(float *out, const float *in,
506 const float *speech_synth,
507 int size, float alpha, float *gain_mem)
510 float speech_energy = 0.0, postfilter_energy = 0.0, gain_scale_factor;
511 float mem = *gain_mem;
513 for (i = 0; i < size; i++) {
514 speech_energy += fabsf(speech_synth[i]);
515 postfilter_energy += fabsf(in[i]);
517 gain_scale_factor = postfilter_energy == 0.0 ? 0.0 :
518 (1.0 - alpha) * speech_energy / postfilter_energy;
520 for (i = 0; i < size; i++) {
521 mem = alpha * mem + gain_scale_factor;
522 out[i] = in[i] * mem;
529 * Kalman smoothing function.
531 * This function looks back pitch +/- 3 samples back into history to find
532 * the best fitting curve (that one giving the optimal gain of the two
533 * signals, i.e. the highest dot product between the two), and then
534 * uses that signal history to smoothen the output of the speech synthesis
537 * @param s WMA Voice decoding context
538 * @param pitch pitch of the speech signal
539 * @param in input speech signal
540 * @param out output pointer for smoothened signal
541 * @param size input/output buffer size
543 * @returns -1 if no smoothening took place, e.g. because no optimal
544 * fit could be found, or 0 on success.
546 static int kalman_smoothen(WMAVoiceContext *s, int pitch,
547 const float *in, float *out, int size)
550 float optimal_gain = 0, dot;
551 const float *ptr = &in[-FFMAX(s->min_pitch_val, pitch - 3)],
552 *end = &in[-FFMIN(s->max_pitch_val, pitch + 3)],
553 *best_hist_ptr = NULL;
555 /* find best fitting point in history */
557 dot = avpriv_scalarproduct_float_c(in, ptr, size);
558 if (dot > optimal_gain) {
562 } while (--ptr >= end);
564 if (optimal_gain <= 0)
566 dot = avpriv_scalarproduct_float_c(best_hist_ptr, best_hist_ptr, size);
567 if (dot <= 0) // would be 1.0
570 if (optimal_gain <= dot) {
571 dot = dot / (dot + 0.6 * optimal_gain); // 0.625-1.000
575 /* actual smoothing */
576 for (n = 0; n < size; n++)
577 out[n] = best_hist_ptr[n] + dot * (in[n] - best_hist_ptr[n]);
583 * Get the tilt factor of a formant filter from its transfer function
584 * @see #tilt_factor() in amrnbdec.c, which does essentially the same,
585 * but somehow (??) it does a speech synthesis filter in the
586 * middle, which is missing here
588 * @param lpcs LPC coefficients
589 * @param n_lpcs Size of LPC buffer
590 * @returns the tilt factor
592 static float tilt_factor(const float *lpcs, int n_lpcs)
596 rh0 = 1.0 + avpriv_scalarproduct_float_c(lpcs, lpcs, n_lpcs);
597 rh1 = lpcs[0] + avpriv_scalarproduct_float_c(lpcs, &lpcs[1], n_lpcs - 1);
603 * Derive denoise filter coefficients (in real domain) from the LPCs.
605 static void calc_input_response(WMAVoiceContext *s, float *lpcs,
606 int fcb_type, float *coeffs, int remainder)
608 float last_coeff, min = 15.0, max = -15.0;
609 float irange, angle_mul, gain_mul, range, sq;
612 /* Create frequency power spectrum of speech input (i.e. RDFT of LPCs) */
613 s->rdft.rdft_calc(&s->rdft, lpcs);
614 #define log_range(var, assign) do { \
615 float tmp = log10f(assign); var = tmp; \
616 max = FFMAX(max, tmp); min = FFMIN(min, tmp); \
618 log_range(last_coeff, lpcs[1] * lpcs[1]);
619 for (n = 1; n < 64; n++)
620 log_range(lpcs[n], lpcs[n * 2] * lpcs[n * 2] +
621 lpcs[n * 2 + 1] * lpcs[n * 2 + 1]);
622 log_range(lpcs[0], lpcs[0] * lpcs[0]);
625 lpcs[64] = last_coeff;
627 /* Now, use this spectrum to pick out these frequencies with higher
628 * (relative) power/energy (which we then take to be "not noise"),
629 * and set up a table (still in lpc[]) of (relative) gains per frequency.
630 * These frequencies will be maintained, while others ("noise") will be
631 * decreased in the filter output. */
632 irange = 64.0 / range; // so irange*(max-value) is in the range [0, 63]
633 gain_mul = range * (fcb_type == FCB_TYPE_HARDCODED ? (5.0 / 13.0) :
635 angle_mul = gain_mul * (8.0 * M_LN10 / M_PI);
636 for (n = 0; n <= 64; n++) {
639 idx = lrint((max - lpcs[n]) * irange - 1);
641 pwr = wmavoice_denoise_power_table[s->denoise_strength][idx];
642 lpcs[n] = angle_mul * pwr;
644 /* 70.57 =~ 1/log10(1.0331663) */
645 idx = av_clipf((pwr * gain_mul - 0.0295) * 70.570526123, 0, INT_MAX / 2);
647 if (idx > 127) { // fall back if index falls outside table range
648 coeffs[n] = wmavoice_energy_table[127] *
649 powf(1.0331663, idx - 127);
651 coeffs[n] = wmavoice_energy_table[FFMAX(0, idx)];
654 /* calculate the Hilbert transform of the gains, which we do (since this
655 * is a sine input) by doing a phase shift (in theory, H(sin())=cos()).
656 * Hilbert_Transform(RDFT(x)) = Laplace_Transform(x), which calculates the
657 * "moment" of the LPCs in this filter. */
658 s->dct.dct_calc(&s->dct, lpcs);
659 s->dst.dct_calc(&s->dst, lpcs);
661 /* Split out the coefficient indexes into phase/magnitude pairs */
662 idx = 255 + av_clip(lpcs[64], -255, 255);
663 coeffs[0] = coeffs[0] * s->cos[idx];
664 idx = 255 + av_clip(lpcs[64] - 2 * lpcs[63], -255, 255);
665 last_coeff = coeffs[64] * s->cos[idx];
667 idx = 255 + av_clip(-lpcs[64] - 2 * lpcs[n - 1], -255, 255);
668 coeffs[n * 2 + 1] = coeffs[n] * s->sin[idx];
669 coeffs[n * 2] = coeffs[n] * s->cos[idx];
673 idx = 255 + av_clip( lpcs[64] - 2 * lpcs[n - 1], -255, 255);
674 coeffs[n * 2 + 1] = coeffs[n] * s->sin[idx];
675 coeffs[n * 2] = coeffs[n] * s->cos[idx];
677 coeffs[1] = last_coeff;
679 /* move into real domain */
680 s->irdft.rdft_calc(&s->irdft, coeffs);
682 /* tilt correction and normalize scale */
683 memset(&coeffs[remainder], 0, sizeof(coeffs[0]) * (128 - remainder));
684 if (s->denoise_tilt_corr) {
687 coeffs[remainder - 1] = 0;
688 ff_tilt_compensation(&tilt_mem,
689 -1.8 * tilt_factor(coeffs, remainder - 1),
692 sq = (1.0 / 64.0) * sqrtf(1 / avpriv_scalarproduct_float_c(coeffs, coeffs,
694 for (n = 0; n < remainder; n++)
699 * This function applies a Wiener filter on the (noisy) speech signal as
700 * a means to denoise it.
702 * - take RDFT of LPCs to get the power spectrum of the noise + speech;
703 * - using this power spectrum, calculate (for each frequency) the Wiener
704 * filter gain, which depends on the frequency power and desired level
705 * of noise subtraction (when set too high, this leads to artifacts)
706 * We can do this symmetrically over the X-axis (so 0-4kHz is the inverse
708 * - by doing a phase shift, calculate the Hilbert transform of this array
709 * of per-frequency filter-gains to get the filtering coefficients;
710 * - smoothen/normalize/de-tilt these filter coefficients as desired;
711 * - take RDFT of noisy sound, apply the coefficients and take its IRDFT
712 * to get the denoised speech signal;
713 * - the leftover (i.e. output of the IRDFT on denoised speech data beyond
714 * the frame boundary) are saved and applied to subsequent frames by an
715 * overlap-add method (otherwise you get clicking-artifacts).
717 * @param s WMA Voice decoding context
718 * @param fcb_type Frame (codebook) type
719 * @param synth_pf input: the noisy speech signal, output: denoised speech
720 * data; should be 16-byte aligned (for ASM purposes)
721 * @param size size of the speech data
722 * @param lpcs LPCs used to synthesize this frame's speech data
724 static void wiener_denoise(WMAVoiceContext *s, int fcb_type,
725 float *synth_pf, int size,
728 int remainder, lim, n;
730 if (fcb_type != FCB_TYPE_SILENCE) {
731 float *tilted_lpcs = s->tilted_lpcs_pf,
732 *coeffs = s->denoise_coeffs_pf, tilt_mem = 0;
734 tilted_lpcs[0] = 1.0;
735 memcpy(&tilted_lpcs[1], lpcs, sizeof(lpcs[0]) * s->lsps);
736 memset(&tilted_lpcs[s->lsps + 1], 0,
737 sizeof(tilted_lpcs[0]) * (128 - s->lsps - 1));
738 ff_tilt_compensation(&tilt_mem, 0.7 * tilt_factor(lpcs, s->lsps),
739 tilted_lpcs, s->lsps + 2);
741 /* The IRDFT output (127 samples for 7-bit filter) beyond the frame
742 * size is applied to the next frame. All input beyond this is zero,
743 * and thus all output beyond this will go towards zero, hence we can
744 * limit to min(size-1, 127-size) as a performance consideration. */
745 remainder = FFMIN(127 - size, size - 1);
746 calc_input_response(s, tilted_lpcs, fcb_type, coeffs, remainder);
748 /* apply coefficients (in frequency spectrum domain), i.e. complex
749 * number multiplication */
750 memset(&synth_pf[size], 0, sizeof(synth_pf[0]) * (128 - size));
751 s->rdft.rdft_calc(&s->rdft, synth_pf);
752 s->rdft.rdft_calc(&s->rdft, coeffs);
753 synth_pf[0] *= coeffs[0];
754 synth_pf[1] *= coeffs[1];
755 for (n = 1; n < 64; n++) {
756 float v1 = synth_pf[n * 2], v2 = synth_pf[n * 2 + 1];
757 synth_pf[n * 2] = v1 * coeffs[n * 2] - v2 * coeffs[n * 2 + 1];
758 synth_pf[n * 2 + 1] = v2 * coeffs[n * 2] + v1 * coeffs[n * 2 + 1];
760 s->irdft.rdft_calc(&s->irdft, synth_pf);
763 /* merge filter output with the history of previous runs */
764 if (s->denoise_filter_cache_size) {
765 lim = FFMIN(s->denoise_filter_cache_size, size);
766 for (n = 0; n < lim; n++)
767 synth_pf[n] += s->denoise_filter_cache[n];
768 s->denoise_filter_cache_size -= lim;
769 memmove(s->denoise_filter_cache, &s->denoise_filter_cache[size],
770 sizeof(s->denoise_filter_cache[0]) * s->denoise_filter_cache_size);
773 /* move remainder of filter output into a cache for future runs */
774 if (fcb_type != FCB_TYPE_SILENCE) {
775 lim = FFMIN(remainder, s->denoise_filter_cache_size);
776 for (n = 0; n < lim; n++)
777 s->denoise_filter_cache[n] += synth_pf[size + n];
778 if (lim < remainder) {
779 memcpy(&s->denoise_filter_cache[lim], &synth_pf[size + lim],
780 sizeof(s->denoise_filter_cache[0]) * (remainder - lim));
781 s->denoise_filter_cache_size = remainder;
787 * Averaging projection filter, the postfilter used in WMAVoice.
789 * This uses the following steps:
790 * - A zero-synthesis filter (generate excitation from synth signal)
791 * - Kalman smoothing on excitation, based on pitch
792 * - Re-synthesized smoothened output
793 * - Iterative Wiener denoise filter
794 * - Adaptive gain filter
797 * @param s WMAVoice decoding context
798 * @param synth Speech synthesis output (before postfilter)
799 * @param samples Output buffer for filtered samples
800 * @param size Buffer size of synth & samples
801 * @param lpcs Generated LPCs used for speech synthesis
802 * @param zero_exc_pf destination for zero synthesis filter (16-byte aligned)
803 * @param fcb_type Frame type (silence, hardcoded, AW-pulses or FCB-pulses)
804 * @param pitch Pitch of the input signal
806 static void postfilter(WMAVoiceContext *s, const float *synth,
807 float *samples, int size,
808 const float *lpcs, float *zero_exc_pf,
809 int fcb_type, int pitch)
811 float synth_filter_in_buf[MAX_FRAMESIZE / 2],
812 *synth_pf = &s->synth_filter_out_buf[MAX_LSPS_ALIGN16],
813 *synth_filter_in = zero_exc_pf;
815 av_assert0(size <= MAX_FRAMESIZE / 2);
817 /* generate excitation from input signal */
818 ff_celp_lp_zero_synthesis_filterf(zero_exc_pf, lpcs, synth, size, s->lsps);
820 if (fcb_type >= FCB_TYPE_AW_PULSES &&
821 !kalman_smoothen(s, pitch, zero_exc_pf, synth_filter_in_buf, size))
822 synth_filter_in = synth_filter_in_buf;
824 /* re-synthesize speech after smoothening, and keep history */
825 ff_celp_lp_synthesis_filterf(synth_pf, lpcs,
826 synth_filter_in, size, s->lsps);
827 memcpy(&synth_pf[-s->lsps], &synth_pf[size - s->lsps],
828 sizeof(synth_pf[0]) * s->lsps);
830 wiener_denoise(s, fcb_type, synth_pf, size, lpcs);
832 adaptive_gain_control(samples, synth_pf, synth, size, 0.99,
835 if (s->dc_level > 8) {
836 /* remove ultra-low frequency DC noise / highpass filter;
837 * coefficients are identical to those used in SIPR decoding,
838 * and very closely resemble those used in AMR-NB decoding. */
839 ff_acelp_apply_order_2_transfer_function(samples, samples,
840 (const float[2]) { -1.99997, 1.0 },
841 (const float[2]) { -1.9330735188, 0.93589198496 },
842 0.93980580475, s->dcf_mem, size);
851 * @param lsps output pointer to the array that will hold the LSPs
852 * @param num number of LSPs to be dequantized
853 * @param values quantized values, contains n_stages values
854 * @param sizes range (i.e. max value) of each quantized value
855 * @param n_stages number of dequantization runs
856 * @param table dequantization table to be used
857 * @param mul_q LSF multiplier
858 * @param base_q base (lowest) LSF values
860 static void dequant_lsps(double *lsps, int num,
861 const uint16_t *values,
862 const uint16_t *sizes,
863 int n_stages, const uint8_t *table,
865 const double *base_q)
869 memset(lsps, 0, num * sizeof(*lsps));
870 for (n = 0; n < n_stages; n++) {
871 const uint8_t *t_off = &table[values[n] * num];
872 double base = base_q[n], mul = mul_q[n];
874 for (m = 0; m < num; m++)
875 lsps[m] += base + mul * t_off[m];
877 table += sizes[n] * num;
882 * @name LSP dequantization routines
883 * LSP dequantization routines, for 10/16LSPs and independent/residual coding.
884 * lsp10i() consumes 24 bits; lsp10r() consumes an additional 24 bits;
885 * lsp16i() consumes 34 bits; lsp16r() consumes an additional 26 bits.
889 * Parse 10 independently-coded LSPs.
891 static void dequant_lsp10i(GetBitContext *gb, double *lsps)
893 static const uint16_t vec_sizes[4] = { 256, 64, 32, 32 };
894 static const double mul_lsf[4] = {
895 5.2187144800e-3, 1.4626986422e-3,
896 9.6179549166e-4, 1.1325736225e-3
898 static const double base_lsf[4] = {
899 M_PI * -2.15522e-1, M_PI * -6.1646e-2,
900 M_PI * -3.3486e-2, M_PI * -5.7408e-2
904 v[0] = get_bits(gb, 8);
905 v[1] = get_bits(gb, 6);
906 v[2] = get_bits(gb, 5);
907 v[3] = get_bits(gb, 5);
909 dequant_lsps(lsps, 10, v, vec_sizes, 4, wmavoice_dq_lsp10i,
914 * Parse 10 independently-coded LSPs, and then derive the tables to
915 * generate LSPs for the other frames from them (residual coding).
917 static void dequant_lsp10r(GetBitContext *gb,
918 double *i_lsps, const double *old,
919 double *a1, double *a2, int q_mode)
921 static const uint16_t vec_sizes[3] = { 128, 64, 64 };
922 static const double mul_lsf[3] = {
923 2.5807601174e-3, 1.2354460219e-3, 1.1763821673e-3
925 static const double base_lsf[3] = {
926 M_PI * -1.07448e-1, M_PI * -5.2706e-2, M_PI * -5.1634e-2
928 const float (*ipol_tab)[2][10] = q_mode ?
929 wmavoice_lsp10_intercoeff_b : wmavoice_lsp10_intercoeff_a;
930 uint16_t interpol, v[3];
933 dequant_lsp10i(gb, i_lsps);
935 interpol = get_bits(gb, 5);
936 v[0] = get_bits(gb, 7);
937 v[1] = get_bits(gb, 6);
938 v[2] = get_bits(gb, 6);
940 for (n = 0; n < 10; n++) {
941 double delta = old[n] - i_lsps[n];
942 a1[n] = ipol_tab[interpol][0][n] * delta + i_lsps[n];
943 a1[10 + n] = ipol_tab[interpol][1][n] * delta + i_lsps[n];
946 dequant_lsps(a2, 20, v, vec_sizes, 3, wmavoice_dq_lsp10r,
951 * Parse 16 independently-coded LSPs.
953 static void dequant_lsp16i(GetBitContext *gb, double *lsps)
955 static const uint16_t vec_sizes[5] = { 256, 64, 128, 64, 128 };
956 static const double mul_lsf[5] = {
957 3.3439586280e-3, 6.9908173703e-4,
958 3.3216608306e-3, 1.0334960326e-3,
961 static const double base_lsf[5] = {
962 M_PI * -1.27576e-1, M_PI * -2.4292e-2,
963 M_PI * -1.28094e-1, M_PI * -3.2128e-2,
968 v[0] = get_bits(gb, 8);
969 v[1] = get_bits(gb, 6);
970 v[2] = get_bits(gb, 7);
971 v[3] = get_bits(gb, 6);
972 v[4] = get_bits(gb, 7);
974 dequant_lsps( lsps, 5, v, vec_sizes, 2,
975 wmavoice_dq_lsp16i1, mul_lsf, base_lsf);
976 dequant_lsps(&lsps[5], 5, &v[2], &vec_sizes[2], 2,
977 wmavoice_dq_lsp16i2, &mul_lsf[2], &base_lsf[2]);
978 dequant_lsps(&lsps[10], 6, &v[4], &vec_sizes[4], 1,
979 wmavoice_dq_lsp16i3, &mul_lsf[4], &base_lsf[4]);
983 * Parse 16 independently-coded LSPs, and then derive the tables to
984 * generate LSPs for the other frames from them (residual coding).
986 static void dequant_lsp16r(GetBitContext *gb,
987 double *i_lsps, const double *old,
988 double *a1, double *a2, int q_mode)
990 static const uint16_t vec_sizes[3] = { 128, 128, 128 };
991 static const double mul_lsf[3] = {
992 1.2232979501e-3, 1.4062241527e-3, 1.6114744851e-3
994 static const double base_lsf[3] = {
995 M_PI * -5.5830e-2, M_PI * -5.2908e-2, M_PI * -5.4776e-2
997 const float (*ipol_tab)[2][16] = q_mode ?
998 wmavoice_lsp16_intercoeff_b : wmavoice_lsp16_intercoeff_a;
999 uint16_t interpol, v[3];
1002 dequant_lsp16i(gb, i_lsps);
1004 interpol = get_bits(gb, 5);
1005 v[0] = get_bits(gb, 7);
1006 v[1] = get_bits(gb, 7);
1007 v[2] = get_bits(gb, 7);
1009 for (n = 0; n < 16; n++) {
1010 double delta = old[n] - i_lsps[n];
1011 a1[n] = ipol_tab[interpol][0][n] * delta + i_lsps[n];
1012 a1[16 + n] = ipol_tab[interpol][1][n] * delta + i_lsps[n];
1015 dequant_lsps( a2, 10, v, vec_sizes, 1,
1016 wmavoice_dq_lsp16r1, mul_lsf, base_lsf);
1017 dequant_lsps(&a2[10], 10, &v[1], &vec_sizes[1], 1,
1018 wmavoice_dq_lsp16r2, &mul_lsf[1], &base_lsf[1]);
1019 dequant_lsps(&a2[20], 12, &v[2], &vec_sizes[2], 1,
1020 wmavoice_dq_lsp16r3, &mul_lsf[2], &base_lsf[2]);
1025 * @name Pitch-adaptive window coding functions
1026 * The next few functions are for pitch-adaptive window coding.
1030 * Parse the offset of the first pitch-adaptive window pulses, and
1031 * the distribution of pulses between the two blocks in this frame.
1032 * @param s WMA Voice decoding context private data
1033 * @param gb bit I/O context
1034 * @param pitch pitch for each block in this frame
1036 static void aw_parse_coords(WMAVoiceContext *s, GetBitContext *gb,
1039 static const int16_t start_offset[94] = {
1040 -11, -9, -7, -5, -3, -1, 1, 3, 5, 7, 9, 11,
1041 13, 15, 18, 17, 19, 20, 21, 22, 23, 24, 25, 26,
1042 27, 28, 29, 30, 31, 32, 33, 35, 37, 39, 41, 43,
1043 45, 47, 49, 51, 53, 55, 57, 59, 61, 63, 65, 67,
1044 69, 71, 73, 75, 77, 79, 81, 83, 85, 87, 89, 91,
1045 93, 95, 97, 99, 101, 103, 105, 107, 109, 111, 113, 115,
1046 117, 119, 121, 123, 125, 127, 129, 131, 133, 135, 137, 139,
1047 141, 143, 145, 147, 149, 151, 153, 155, 157, 159
1051 /* position of pulse */
1052 s->aw_idx_is_ext = 0;
1053 if ((bits = get_bits(gb, 6)) >= 54) {
1054 s->aw_idx_is_ext = 1;
1055 bits += (bits - 54) * 3 + get_bits(gb, 2);
1058 /* for a repeated pulse at pulse_off with a pitch_lag of pitch[], count
1059 * the distribution of the pulses in each block contained in this frame. */
1060 s->aw_pulse_range = FFMIN(pitch[0], pitch[1]) > 32 ? 24 : 16;
1061 for (offset = start_offset[bits]; offset < 0; offset += pitch[0]) ;
1062 s->aw_n_pulses[0] = (pitch[0] - 1 + MAX_FRAMESIZE / 2 - offset) / pitch[0];
1063 s->aw_first_pulse_off[0] = offset - s->aw_pulse_range / 2;
1064 offset += s->aw_n_pulses[0] * pitch[0];
1065 s->aw_n_pulses[1] = (pitch[1] - 1 + MAX_FRAMESIZE - offset) / pitch[1];
1066 s->aw_first_pulse_off[1] = offset - (MAX_FRAMESIZE + s->aw_pulse_range) / 2;
1068 /* if continuing from a position before the block, reset position to
1069 * start of block (when corrected for the range over which it can be
1070 * spread in aw_pulse_set1()). */
1071 if (start_offset[bits] < MAX_FRAMESIZE / 2) {
1072 while (s->aw_first_pulse_off[1] - pitch[1] + s->aw_pulse_range > 0)
1073 s->aw_first_pulse_off[1] -= pitch[1];
1074 if (start_offset[bits] < 0)
1075 while (s->aw_first_pulse_off[0] - pitch[0] + s->aw_pulse_range > 0)
1076 s->aw_first_pulse_off[0] -= pitch[0];
1081 * Apply second set of pitch-adaptive window pulses.
1082 * @param s WMA Voice decoding context private data
1083 * @param gb bit I/O context
1084 * @param block_idx block index in frame [0, 1]
1085 * @param fcb structure containing fixed codebook vector info
1086 * @return -1 on error, 0 otherwise
1088 static int aw_pulse_set2(WMAVoiceContext *s, GetBitContext *gb,
1089 int block_idx, AMRFixed *fcb)
1091 uint16_t use_mask_mem[9]; // only 5 are used, rest is padding
1092 uint16_t *use_mask = use_mask_mem + 2;
1093 /* in this function, idx is the index in the 80-bit (+ padding) use_mask
1094 * bit-array. Since use_mask consists of 16-bit values, the lower 4 bits
1095 * of idx are the position of the bit within a particular item in the
1096 * array (0 being the most significant bit, and 15 being the least
1097 * significant bit), and the remainder (>> 4) is the index in the
1098 * use_mask[]-array. This is faster and uses less memory than using a
1099 * 80-byte/80-int array. */
1100 int pulse_off = s->aw_first_pulse_off[block_idx],
1101 pulse_start, n, idx, range, aidx, start_off = 0;
1103 /* set offset of first pulse to within this block */
1104 if (s->aw_n_pulses[block_idx] > 0)
1105 while (pulse_off + s->aw_pulse_range < 1)
1106 pulse_off += fcb->pitch_lag;
1108 /* find range per pulse */
1109 if (s->aw_n_pulses[0] > 0) {
1110 if (block_idx == 0) {
1112 } else /* block_idx = 1 */ {
1114 if (s->aw_n_pulses[block_idx] > 0)
1115 pulse_off = s->aw_next_pulse_off_cache;
1119 pulse_start = s->aw_n_pulses[block_idx] > 0 ? pulse_off - range / 2 : 0;
1121 /* aw_pulse_set1() already applies pulses around pulse_off (to be exactly,
1122 * in the range of [pulse_off, pulse_off + s->aw_pulse_range], and thus
1123 * we exclude that range from being pulsed again in this function. */
1124 memset(&use_mask[-2], 0, 2 * sizeof(use_mask[0]));
1125 memset( use_mask, -1, 5 * sizeof(use_mask[0]));
1126 memset(&use_mask[5], 0, 2 * sizeof(use_mask[0]));
1127 if (s->aw_n_pulses[block_idx] > 0)
1128 for (idx = pulse_off; idx < MAX_FRAMESIZE / 2; idx += fcb->pitch_lag) {
1129 int excl_range = s->aw_pulse_range; // always 16 or 24
1130 uint16_t *use_mask_ptr = &use_mask[idx >> 4];
1131 int first_sh = 16 - (idx & 15);
1132 *use_mask_ptr++ &= 0xFFFFu << first_sh;
1133 excl_range -= first_sh;
1134 if (excl_range >= 16) {
1135 *use_mask_ptr++ = 0;
1136 *use_mask_ptr &= 0xFFFF >> (excl_range - 16);
1138 *use_mask_ptr &= 0xFFFF >> excl_range;
1141 /* find the 'aidx'th offset that is not excluded */
1142 aidx = get_bits(gb, s->aw_n_pulses[0] > 0 ? 5 - 2 * block_idx : 4);
1143 for (n = 0; n <= aidx; pulse_start++) {
1144 for (idx = pulse_start; idx < 0; idx += fcb->pitch_lag) ;
1145 if (idx >= MAX_FRAMESIZE / 2) { // find from zero
1146 if (use_mask[0]) idx = 0x0F;
1147 else if (use_mask[1]) idx = 0x1F;
1148 else if (use_mask[2]) idx = 0x2F;
1149 else if (use_mask[3]) idx = 0x3F;
1150 else if (use_mask[4]) idx = 0x4F;
1152 idx -= av_log2_16bit(use_mask[idx >> 4]);
1154 if (use_mask[idx >> 4] & (0x8000 >> (idx & 15))) {
1155 use_mask[idx >> 4] &= ~(0x8000 >> (idx & 15));
1161 fcb->x[fcb->n] = start_off;
1162 fcb->y[fcb->n] = get_bits1(gb) ? -1.0 : 1.0;
1165 /* set offset for next block, relative to start of that block */
1166 n = (MAX_FRAMESIZE / 2 - start_off) % fcb->pitch_lag;
1167 s->aw_next_pulse_off_cache = n ? fcb->pitch_lag - n : 0;
1172 * Apply first set of pitch-adaptive window pulses.
1173 * @param s WMA Voice decoding context private data
1174 * @param gb bit I/O context
1175 * @param block_idx block index in frame [0, 1]
1176 * @param fcb storage location for fixed codebook pulse info
1178 static void aw_pulse_set1(WMAVoiceContext *s, GetBitContext *gb,
1179 int block_idx, AMRFixed *fcb)
1181 int val = get_bits(gb, 12 - 2 * (s->aw_idx_is_ext && !block_idx));
1184 if (s->aw_n_pulses[block_idx] > 0) {
1185 int n, v_mask, i_mask, sh, n_pulses;
1187 if (s->aw_pulse_range == 24) { // 3 pulses, 1:sign + 3:index each
1192 } else { // 4 pulses, 1:sign + 2:index each
1199 for (n = n_pulses - 1; n >= 0; n--, val >>= sh) {
1200 fcb->y[fcb->n] = (val & v_mask) ? -1.0 : 1.0;
1201 fcb->x[fcb->n] = (val & i_mask) * n_pulses + n +
1202 s->aw_first_pulse_off[block_idx];
1203 while (fcb->x[fcb->n] < 0)
1204 fcb->x[fcb->n] += fcb->pitch_lag;
1205 if (fcb->x[fcb->n] < MAX_FRAMESIZE / 2)
1209 int num2 = (val & 0x1FF) >> 1, delta, idx;
1211 if (num2 < 1 * 79) { delta = 1; idx = num2 + 1; }
1212 else if (num2 < 2 * 78) { delta = 3; idx = num2 + 1 - 1 * 77; }
1213 else if (num2 < 3 * 77) { delta = 5; idx = num2 + 1 - 2 * 76; }
1214 else { delta = 7; idx = num2 + 1 - 3 * 75; }
1215 v = (val & 0x200) ? -1.0 : 1.0;
1217 fcb->no_repeat_mask |= 3 << fcb->n;
1218 fcb->x[fcb->n] = idx - delta;
1220 fcb->x[fcb->n + 1] = idx;
1221 fcb->y[fcb->n + 1] = (val & 1) ? -v : v;
1229 * Generate a random number from frame_cntr and block_idx, which will live
1230 * in the range [0, 1000 - block_size] (so it can be used as an index in a
1231 * table of size 1000 of which you want to read block_size entries).
1233 * @param frame_cntr current frame number
1234 * @param block_num current block index
1235 * @param block_size amount of entries we want to read from a table
1236 * that has 1000 entries
1237 * @return a (non-)random number in the [0, 1000 - block_size] range.
1239 static int pRNG(int frame_cntr, int block_num, int block_size)
1241 /* array to simplify the calculation of z:
1242 * y = (x % 9) * 5 + 6;
1243 * z = (49995 * x) / y;
1244 * Since y only has 9 values, we can remove the division by using a
1245 * LUT and using FASTDIV-style divisions. For each of the 9 values
1246 * of y, we can rewrite z as:
1247 * z = x * (49995 / y) + x * ((49995 % y) / y)
1248 * In this table, each col represents one possible value of y, the
1249 * first number is 49995 / y, and the second is the FASTDIV variant
1250 * of 49995 % y / y. */
1251 static const unsigned int div_tbl[9][2] = {
1252 { 8332, 3 * 715827883U }, // y = 6
1253 { 4545, 0 * 390451573U }, // y = 11
1254 { 3124, 11 * 268435456U }, // y = 16
1255 { 2380, 15 * 204522253U }, // y = 21
1256 { 1922, 23 * 165191050U }, // y = 26
1257 { 1612, 23 * 138547333U }, // y = 31
1258 { 1388, 27 * 119304648U }, // y = 36
1259 { 1219, 16 * 104755300U }, // y = 41
1260 { 1086, 39 * 93368855U } // y = 46
1262 unsigned int z, y, x = MUL16(block_num, 1877) + frame_cntr;
1263 if (x >= 0xFFFF) x -= 0xFFFF; // max value of x is 8*1877+0xFFFE=0x13AA6,
1264 // so this is effectively a modulo (%)
1265 y = x - 9 * MULH(477218589, x); // x % 9
1266 z = (uint16_t) (x * div_tbl[y][0] + UMULH(x, div_tbl[y][1]));
1267 // z = x * 49995 / (y * 5 + 6)
1268 return z % (1000 - block_size);
1272 * Parse hardcoded signal for a single block.
1273 * @note see #synth_block().
1275 static void synth_block_hardcoded(WMAVoiceContext *s, GetBitContext *gb,
1276 int block_idx, int size,
1277 const struct frame_type_desc *frame_desc,
1283 av_assert0(size <= MAX_FRAMESIZE);
1285 /* Set the offset from which we start reading wmavoice_std_codebook */
1286 if (frame_desc->fcb_type == FCB_TYPE_SILENCE) {
1287 r_idx = pRNG(s->frame_cntr, block_idx, size);
1288 gain = s->silence_gain;
1289 } else /* FCB_TYPE_HARDCODED */ {
1290 r_idx = get_bits(gb, 8);
1291 gain = wmavoice_gain_universal[get_bits(gb, 6)];
1294 /* Clear gain prediction parameters */
1295 memset(s->gain_pred_err, 0, sizeof(s->gain_pred_err));
1297 /* Apply gain to hardcoded codebook and use that as excitation signal */
1298 for (n = 0; n < size; n++)
1299 excitation[n] = wmavoice_std_codebook[r_idx + n] * gain;
1303 * Parse FCB/ACB signal for a single block.
1304 * @note see #synth_block().
1306 static void synth_block_fcb_acb(WMAVoiceContext *s, GetBitContext *gb,
1307 int block_idx, int size,
1308 int block_pitch_sh2,
1309 const struct frame_type_desc *frame_desc,
1312 static const float gain_coeff[6] = {
1313 0.8169, -0.06545, 0.1726, 0.0185, -0.0359, 0.0458
1315 float pulses[MAX_FRAMESIZE / 2], pred_err, acb_gain, fcb_gain;
1316 int n, idx, gain_weight;
1319 av_assert0(size <= MAX_FRAMESIZE / 2);
1320 memset(pulses, 0, sizeof(*pulses) * size);
1322 fcb.pitch_lag = block_pitch_sh2 >> 2;
1323 fcb.pitch_fac = 1.0;
1324 fcb.no_repeat_mask = 0;
1327 /* For the other frame types, this is where we apply the innovation
1328 * (fixed) codebook pulses of the speech signal. */
1329 if (frame_desc->fcb_type == FCB_TYPE_AW_PULSES) {
1330 aw_pulse_set1(s, gb, block_idx, &fcb);
1331 if (aw_pulse_set2(s, gb, block_idx, &fcb)) {
1332 /* Conceal the block with silence and return.
1333 * Skip the correct amount of bits to read the next
1334 * block from the correct offset. */
1335 int r_idx = pRNG(s->frame_cntr, block_idx, size);
1337 for (n = 0; n < size; n++)
1339 wmavoice_std_codebook[r_idx + n] * s->silence_gain;
1340 skip_bits(gb, 7 + 1);
1343 } else /* FCB_TYPE_EXC_PULSES */ {
1344 int offset_nbits = 5 - frame_desc->log_n_blocks;
1346 fcb.no_repeat_mask = -1;
1347 /* similar to ff_decode_10_pulses_35bits(), but with single pulses
1348 * (instead of double) for a subset of pulses */
1349 for (n = 0; n < 5; n++) {
1353 sign = get_bits1(gb) ? 1.0 : -1.0;
1354 pos1 = get_bits(gb, offset_nbits);
1355 fcb.x[fcb.n] = n + 5 * pos1;
1356 fcb.y[fcb.n++] = sign;
1357 if (n < frame_desc->dbl_pulses) {
1358 pos2 = get_bits(gb, offset_nbits);
1359 fcb.x[fcb.n] = n + 5 * pos2;
1360 fcb.y[fcb.n++] = (pos1 < pos2) ? -sign : sign;
1364 ff_set_fixed_vector(pulses, &fcb, 1.0, size);
1366 /* Calculate gain for adaptive & fixed codebook signal.
1367 * see ff_amr_set_fixed_gain(). */
1368 idx = get_bits(gb, 7);
1369 fcb_gain = expf(avpriv_scalarproduct_float_c(s->gain_pred_err,
1371 5.2409161640 + wmavoice_gain_codebook_fcb[idx]);
1372 acb_gain = wmavoice_gain_codebook_acb[idx];
1373 pred_err = av_clipf(wmavoice_gain_codebook_fcb[idx],
1374 -2.9957322736 /* log(0.05) */,
1375 1.6094379124 /* log(5.0) */);
1377 gain_weight = 8 >> frame_desc->log_n_blocks;
1378 memmove(&s->gain_pred_err[gain_weight], s->gain_pred_err,
1379 sizeof(*s->gain_pred_err) * (6 - gain_weight));
1380 for (n = 0; n < gain_weight; n++)
1381 s->gain_pred_err[n] = pred_err;
1383 /* Calculation of adaptive codebook */
1384 if (frame_desc->acb_type == ACB_TYPE_ASYMMETRIC) {
1386 for (n = 0; n < size; n += len) {
1388 int abs_idx = block_idx * size + n;
1389 int pitch_sh16 = (s->last_pitch_val << 16) +
1390 s->pitch_diff_sh16 * abs_idx;
1391 int pitch = (pitch_sh16 + 0x6FFF) >> 16;
1392 int idx_sh16 = ((pitch << 16) - pitch_sh16) * 8 + 0x58000;
1393 idx = idx_sh16 >> 16;
1394 if (s->pitch_diff_sh16) {
1395 if (s->pitch_diff_sh16 > 0) {
1396 next_idx_sh16 = (idx_sh16) &~ 0xFFFF;
1398 next_idx_sh16 = (idx_sh16 + 0x10000) &~ 0xFFFF;
1399 len = av_clip((idx_sh16 - next_idx_sh16) / s->pitch_diff_sh16 / 8,
1404 ff_acelp_interpolatef(&excitation[n], &excitation[n - pitch],
1405 wmavoice_ipol1_coeffs, 17,
1408 } else /* ACB_TYPE_HAMMING */ {
1409 int block_pitch = block_pitch_sh2 >> 2;
1410 idx = block_pitch_sh2 & 3;
1412 ff_acelp_interpolatef(excitation, &excitation[-block_pitch],
1413 wmavoice_ipol2_coeffs, 4,
1416 av_memcpy_backptr((uint8_t *) excitation, sizeof(float) * block_pitch,
1417 sizeof(float) * size);
1420 /* Interpolate ACB/FCB and use as excitation signal */
1421 ff_weighted_vector_sumf(excitation, excitation, pulses,
1422 acb_gain, fcb_gain, size);
1426 * Parse data in a single block.
1428 * @param s WMA Voice decoding context private data
1429 * @param gb bit I/O context
1430 * @param block_idx index of the to-be-read block
1431 * @param size amount of samples to be read in this block
1432 * @param block_pitch_sh2 pitch for this block << 2
1433 * @param lsps LSPs for (the end of) this frame
1434 * @param prev_lsps LSPs for the last frame
1435 * @param frame_desc frame type descriptor
1436 * @param excitation target memory for the ACB+FCB interpolated signal
1437 * @param synth target memory for the speech synthesis filter output
1438 * @return 0 on success, <0 on error.
1440 static void synth_block(WMAVoiceContext *s, GetBitContext *gb,
1441 int block_idx, int size,
1442 int block_pitch_sh2,
1443 const double *lsps, const double *prev_lsps,
1444 const struct frame_type_desc *frame_desc,
1445 float *excitation, float *synth)
1447 double i_lsps[MAX_LSPS];
1448 float lpcs[MAX_LSPS];
1452 if (frame_desc->acb_type == ACB_TYPE_NONE)
1453 synth_block_hardcoded(s, gb, block_idx, size, frame_desc, excitation);
1455 synth_block_fcb_acb(s, gb, block_idx, size, block_pitch_sh2,
1456 frame_desc, excitation);
1458 /* convert interpolated LSPs to LPCs */
1459 fac = (block_idx + 0.5) / frame_desc->n_blocks;
1460 for (n = 0; n < s->lsps; n++) // LSF -> LSP
1461 i_lsps[n] = cos(prev_lsps[n] + fac * (lsps[n] - prev_lsps[n]));
1462 ff_acelp_lspd2lpc(i_lsps, lpcs, s->lsps >> 1);
1464 /* Speech synthesis */
1465 ff_celp_lp_synthesis_filterf(synth, lpcs, excitation, size, s->lsps);
1469 * Synthesize output samples for a single frame.
1471 * @param ctx WMA Voice decoder context
1472 * @param gb bit I/O context (s->gb or one for cross-packet superframes)
1473 * @param frame_idx Frame number within superframe [0-2]
1474 * @param samples pointer to output sample buffer, has space for at least 160
1476 * @param lsps LSP array
1477 * @param prev_lsps array of previous frame's LSPs
1478 * @param excitation target buffer for excitation signal
1479 * @param synth target buffer for synthesized speech data
1480 * @return 0 on success, <0 on error.
1482 static int synth_frame(AVCodecContext *ctx, GetBitContext *gb, int frame_idx,
1484 const double *lsps, const double *prev_lsps,
1485 float *excitation, float *synth)
1487 WMAVoiceContext *s = ctx->priv_data;
1488 int n, n_blocks_x2, log_n_blocks_x2, av_uninit(cur_pitch_val);
1489 int pitch[MAX_BLOCKS], av_uninit(last_block_pitch);
1491 /* Parse frame type ("frame header"), see frame_descs */
1492 int bd_idx = s->vbm_tree[get_vlc2(gb, frame_type_vlc.table, 6, 3)], block_nsamples;
1495 av_log(ctx, AV_LOG_ERROR,
1496 "Invalid frame type VLC code, skipping\n");
1497 return AVERROR_INVALIDDATA;
1500 block_nsamples = MAX_FRAMESIZE / frame_descs[bd_idx].n_blocks;
1502 /* Pitch calculation for ACB_TYPE_ASYMMETRIC ("pitch-per-frame") */
1503 if (frame_descs[bd_idx].acb_type == ACB_TYPE_ASYMMETRIC) {
1504 /* Pitch is provided per frame, which is interpreted as the pitch of
1505 * the last sample of the last block of this frame. We can interpolate
1506 * the pitch of other blocks (and even pitch-per-sample) by gradually
1507 * incrementing/decrementing prev_frame_pitch to cur_pitch_val. */
1508 n_blocks_x2 = frame_descs[bd_idx].n_blocks << 1;
1509 log_n_blocks_x2 = frame_descs[bd_idx].log_n_blocks + 1;
1510 cur_pitch_val = s->min_pitch_val + get_bits(gb, s->pitch_nbits);
1511 cur_pitch_val = FFMIN(cur_pitch_val, s->max_pitch_val - 1);
1512 if (s->last_acb_type == ACB_TYPE_NONE ||
1513 20 * abs(cur_pitch_val - s->last_pitch_val) >
1514 (cur_pitch_val + s->last_pitch_val))
1515 s->last_pitch_val = cur_pitch_val;
1517 /* pitch per block */
1518 for (n = 0; n < frame_descs[bd_idx].n_blocks; n++) {
1519 int fac = n * 2 + 1;
1521 pitch[n] = (MUL16(fac, cur_pitch_val) +
1522 MUL16((n_blocks_x2 - fac), s->last_pitch_val) +
1523 frame_descs[bd_idx].n_blocks) >> log_n_blocks_x2;
1526 /* "pitch-diff-per-sample" for calculation of pitch per sample */
1527 s->pitch_diff_sh16 =
1528 (cur_pitch_val - s->last_pitch_val) * (1 << 16) / MAX_FRAMESIZE;
1531 /* Global gain (if silence) and pitch-adaptive window coordinates */
1532 switch (frame_descs[bd_idx].fcb_type) {
1533 case FCB_TYPE_SILENCE:
1534 s->silence_gain = wmavoice_gain_silence[get_bits(gb, 8)];
1536 case FCB_TYPE_AW_PULSES:
1537 aw_parse_coords(s, gb, pitch);
1541 for (n = 0; n < frame_descs[bd_idx].n_blocks; n++) {
1544 /* Pitch calculation for ACB_TYPE_HAMMING ("pitch-per-block") */
1545 switch (frame_descs[bd_idx].acb_type) {
1546 case ACB_TYPE_HAMMING: {
1547 /* Pitch is given per block. Per-block pitches are encoded as an
1548 * absolute value for the first block, and then delta values
1549 * relative to this value) for all subsequent blocks. The scale of
1550 * this pitch value is semi-logarithmic compared to its use in the
1551 * decoder, so we convert it to normal scale also. */
1553 t1 = (s->block_conv_table[1] - s->block_conv_table[0]) << 2,
1554 t2 = (s->block_conv_table[2] - s->block_conv_table[1]) << 1,
1555 t3 = s->block_conv_table[3] - s->block_conv_table[2] + 1;
1558 block_pitch = get_bits(gb, s->block_pitch_nbits);
1560 block_pitch = last_block_pitch - s->block_delta_pitch_hrange +
1561 get_bits(gb, s->block_delta_pitch_nbits);
1562 /* Convert last_ so that any next delta is within _range */
1563 last_block_pitch = av_clip(block_pitch,
1564 s->block_delta_pitch_hrange,
1565 s->block_pitch_range -
1566 s->block_delta_pitch_hrange);
1568 /* Convert semi-log-style scale back to normal scale */
1569 if (block_pitch < t1) {
1570 bl_pitch_sh2 = (s->block_conv_table[0] << 2) + block_pitch;
1573 if (block_pitch < t2) {
1575 (s->block_conv_table[1] << 2) + (block_pitch << 1);
1578 if (block_pitch < t3) {
1580 (s->block_conv_table[2] + block_pitch) << 2;
1582 bl_pitch_sh2 = s->block_conv_table[3] << 2;
1585 pitch[n] = bl_pitch_sh2 >> 2;
1589 case ACB_TYPE_ASYMMETRIC: {
1590 bl_pitch_sh2 = pitch[n] << 2;
1594 default: // ACB_TYPE_NONE has no pitch
1599 synth_block(s, gb, n, block_nsamples, bl_pitch_sh2,
1600 lsps, prev_lsps, &frame_descs[bd_idx],
1601 &excitation[n * block_nsamples],
1602 &synth[n * block_nsamples]);
1605 /* Averaging projection filter, if applicable. Else, just copy samples
1606 * from synthesis buffer */
1608 double i_lsps[MAX_LSPS];
1609 float lpcs[MAX_LSPS];
1611 for (n = 0; n < s->lsps; n++) // LSF -> LSP
1612 i_lsps[n] = cos(0.5 * (prev_lsps[n] + lsps[n]));
1613 ff_acelp_lspd2lpc(i_lsps, lpcs, s->lsps >> 1);
1614 postfilter(s, synth, samples, 80, lpcs,
1615 &s->zero_exc_pf[s->history_nsamples + MAX_FRAMESIZE * frame_idx],
1616 frame_descs[bd_idx].fcb_type, pitch[0]);
1618 for (n = 0; n < s->lsps; n++) // LSF -> LSP
1619 i_lsps[n] = cos(lsps[n]);
1620 ff_acelp_lspd2lpc(i_lsps, lpcs, s->lsps >> 1);
1621 postfilter(s, &synth[80], &samples[80], 80, lpcs,
1622 &s->zero_exc_pf[s->history_nsamples + MAX_FRAMESIZE * frame_idx + 80],
1623 frame_descs[bd_idx].fcb_type, pitch[0]);
1625 memcpy(samples, synth, 160 * sizeof(synth[0]));
1627 /* Cache values for next frame */
1629 if (s->frame_cntr >= 0xFFFF) s->frame_cntr -= 0xFFFF; // i.e. modulo (%)
1630 s->last_acb_type = frame_descs[bd_idx].acb_type;
1631 switch (frame_descs[bd_idx].acb_type) {
1633 s->last_pitch_val = 0;
1635 case ACB_TYPE_ASYMMETRIC:
1636 s->last_pitch_val = cur_pitch_val;
1638 case ACB_TYPE_HAMMING:
1639 s->last_pitch_val = pitch[frame_descs[bd_idx].n_blocks - 1];
1647 * Ensure minimum value for first item, maximum value for last value,
1648 * proper spacing between each value and proper ordering.
1650 * @param lsps array of LSPs
1651 * @param num size of LSP array
1653 * @note basically a double version of #ff_acelp_reorder_lsf(), might be
1654 * useful to put in a generic location later on. Parts are also
1655 * present in #ff_set_min_dist_lsf() + #ff_sort_nearly_sorted_floats(),
1656 * which is in float.
1658 static void stabilize_lsps(double *lsps, int num)
1662 /* set minimum value for first, maximum value for last and minimum
1663 * spacing between LSF values.
1664 * Very similar to ff_set_min_dist_lsf(), but in double. */
1665 lsps[0] = FFMAX(lsps[0], 0.0015 * M_PI);
1666 for (n = 1; n < num; n++)
1667 lsps[n] = FFMAX(lsps[n], lsps[n - 1] + 0.0125 * M_PI);
1668 lsps[num - 1] = FFMIN(lsps[num - 1], 0.9985 * M_PI);
1670 /* reorder (looks like one-time / non-recursed bubblesort).
1671 * Very similar to ff_sort_nearly_sorted_floats(), but in double. */
1672 for (n = 1; n < num; n++) {
1673 if (lsps[n] < lsps[n - 1]) {
1674 for (m = 1; m < num; m++) {
1675 double tmp = lsps[m];
1676 for (l = m - 1; l >= 0; l--) {
1677 if (lsps[l] <= tmp) break;
1678 lsps[l + 1] = lsps[l];
1688 * Synthesize output samples for a single superframe. If we have any data
1689 * cached in s->sframe_cache, that will be used instead of whatever is loaded
1692 * WMA Voice superframes contain 3 frames, each containing 160 audio samples,
1693 * to give a total of 480 samples per frame. See #synth_frame() for frame
1694 * parsing. In addition to 3 frames, superframes can also contain the LSPs
1695 * (if these are globally specified for all frames (residually); they can
1696 * also be specified individually per-frame. See the s->has_residual_lsps
1697 * option), and can specify the number of samples encoded in this superframe
1698 * (if less than 480), usually used to prevent blanks at track boundaries.
1700 * @param ctx WMA Voice decoder context
1701 * @return 0 on success, <0 on error or 1 if there was not enough data to
1702 * fully parse the superframe
1704 static int synth_superframe(AVCodecContext *ctx, AVFrame *frame,
1707 WMAVoiceContext *s = ctx->priv_data;
1708 GetBitContext *gb = &s->gb, s_gb;
1709 int n, res, n_samples = MAX_SFRAMESIZE;
1710 double lsps[MAX_FRAMES][MAX_LSPS];
1711 const double *mean_lsf = s->lsps == 16 ?
1712 wmavoice_mean_lsf16[s->lsp_def_mode] : wmavoice_mean_lsf10[s->lsp_def_mode];
1713 float excitation[MAX_SIGNAL_HISTORY + MAX_SFRAMESIZE + 12];
1714 float synth[MAX_LSPS + MAX_SFRAMESIZE];
1717 memcpy(synth, s->synth_history,
1718 s->lsps * sizeof(*synth));
1719 memcpy(excitation, s->excitation_history,
1720 s->history_nsamples * sizeof(*excitation));
1722 if (s->sframe_cache_size > 0) {
1724 init_get_bits(gb, s->sframe_cache, s->sframe_cache_size);
1725 s->sframe_cache_size = 0;
1728 /* First bit is speech/music bit, it differentiates between WMAVoice
1729 * speech samples (the actual codec) and WMAVoice music samples, which
1730 * are really WMAPro-in-WMAVoice-superframes. I've never seen those in
1732 if (!get_bits1(gb)) {
1733 avpriv_request_sample(ctx, "WMAPro-in-WMAVoice");
1734 return AVERROR_PATCHWELCOME;
1737 /* (optional) nr. of samples in superframe; always <= 480 and >= 0 */
1738 if (get_bits1(gb)) {
1739 if ((n_samples = get_bits(gb, 12)) > MAX_SFRAMESIZE) {
1740 av_log(ctx, AV_LOG_ERROR,
1741 "Superframe encodes > %d samples (%d), not allowed\n",
1742 MAX_SFRAMESIZE, n_samples);
1743 return AVERROR_INVALIDDATA;
1747 /* Parse LSPs, if global for the superframe (can also be per-frame). */
1748 if (s->has_residual_lsps) {
1749 double prev_lsps[MAX_LSPS], a1[MAX_LSPS * 2], a2[MAX_LSPS * 2];
1751 for (n = 0; n < s->lsps; n++)
1752 prev_lsps[n] = s->prev_lsps[n] - mean_lsf[n];
1754 if (s->lsps == 10) {
1755 dequant_lsp10r(gb, lsps[2], prev_lsps, a1, a2, s->lsp_q_mode);
1756 } else /* s->lsps == 16 */
1757 dequant_lsp16r(gb, lsps[2], prev_lsps, a1, a2, s->lsp_q_mode);
1759 for (n = 0; n < s->lsps; n++) {
1760 lsps[0][n] = mean_lsf[n] + (a1[n] - a2[n * 2]);
1761 lsps[1][n] = mean_lsf[n] + (a1[s->lsps + n] - a2[n * 2 + 1]);
1762 lsps[2][n] += mean_lsf[n];
1764 for (n = 0; n < 3; n++)
1765 stabilize_lsps(lsps[n], s->lsps);
1768 /* synth_superframe can run multiple times per packet
1769 * free potential previous frame */
1770 av_frame_unref(frame);
1772 /* get output buffer */
1773 frame->nb_samples = MAX_SFRAMESIZE;
1774 if ((res = ff_get_buffer(ctx, frame, 0)) < 0)
1776 frame->nb_samples = n_samples;
1777 samples = (float *)frame->data[0];
1779 /* Parse frames, optionally preceded by per-frame (independent) LSPs. */
1780 for (n = 0; n < 3; n++) {
1781 if (!s->has_residual_lsps) {
1784 if (s->lsps == 10) {
1785 dequant_lsp10i(gb, lsps[n]);
1786 } else /* s->lsps == 16 */
1787 dequant_lsp16i(gb, lsps[n]);
1789 for (m = 0; m < s->lsps; m++)
1790 lsps[n][m] += mean_lsf[m];
1791 stabilize_lsps(lsps[n], s->lsps);
1794 if ((res = synth_frame(ctx, gb, n,
1795 &samples[n * MAX_FRAMESIZE],
1796 lsps[n], n == 0 ? s->prev_lsps : lsps[n - 1],
1797 &excitation[s->history_nsamples + n * MAX_FRAMESIZE],
1798 &synth[s->lsps + n * MAX_FRAMESIZE]))) {
1804 /* Statistics? FIXME - we don't check for length, a slight overrun
1805 * will be caught by internal buffer padding, and anything else
1806 * will be skipped, not read. */
1807 if (get_bits1(gb)) {
1808 res = get_bits(gb, 4);
1809 skip_bits(gb, 10 * (res + 1));
1812 if (get_bits_left(gb) < 0) {
1813 wmavoice_flush(ctx);
1814 return AVERROR_INVALIDDATA;
1819 /* Update history */
1820 memcpy(s->prev_lsps, lsps[2],
1821 s->lsps * sizeof(*s->prev_lsps));
1822 memcpy(s->synth_history, &synth[MAX_SFRAMESIZE],
1823 s->lsps * sizeof(*synth));
1824 memcpy(s->excitation_history, &excitation[MAX_SFRAMESIZE],
1825 s->history_nsamples * sizeof(*excitation));
1827 memmove(s->zero_exc_pf, &s->zero_exc_pf[MAX_SFRAMESIZE],
1828 s->history_nsamples * sizeof(*s->zero_exc_pf));
1834 * Parse the packet header at the start of each packet (input data to this
1837 * @param s WMA Voice decoding context private data
1838 * @return <0 on error, nb_superframes on success.
1840 static int parse_packet_header(WMAVoiceContext *s)
1842 GetBitContext *gb = &s->gb;
1843 unsigned int res, n_superframes = 0;
1845 skip_bits(gb, 4); // packet sequence number
1846 s->has_residual_lsps = get_bits1(gb);
1848 if (get_bits_left(gb) < 6 + s->spillover_bitsize)
1849 return AVERROR_INVALIDDATA;
1851 res = get_bits(gb, 6); // number of superframes per packet
1852 // (minus first one if there is spillover)
1853 n_superframes += res;
1854 } while (res == 0x3F);
1855 s->spillover_nbits = get_bits(gb, s->spillover_bitsize);
1857 return get_bits_left(gb) >= 0 ? n_superframes : AVERROR_INVALIDDATA;
1861 * Copy (unaligned) bits from gb/data/size to pb.
1863 * @param pb target buffer to copy bits into
1864 * @param data source buffer to copy bits from
1865 * @param size size of the source data, in bytes
1866 * @param gb bit I/O context specifying the current position in the source.
1867 * data. This function might use this to align the bit position to
1868 * a whole-byte boundary before calling #ff_copy_bits() on aligned
1870 * @param nbits the amount of bits to copy from source to target
1872 * @note after calling this function, the current position in the input bit
1873 * I/O context is undefined.
1875 static void copy_bits(PutBitContext *pb,
1876 const uint8_t *data, int size,
1877 GetBitContext *gb, int nbits)
1879 int rmn_bytes, rmn_bits;
1881 rmn_bits = rmn_bytes = get_bits_left(gb);
1882 if (rmn_bits < nbits)
1884 if (nbits > pb->size_in_bits - put_bits_count(pb))
1886 rmn_bits &= 7; rmn_bytes >>= 3;
1887 if ((rmn_bits = FFMIN(rmn_bits, nbits)) > 0)
1888 put_bits(pb, rmn_bits, get_bits(gb, rmn_bits));
1889 ff_copy_bits(pb, data + size - rmn_bytes,
1890 FFMIN(nbits - rmn_bits, rmn_bytes << 3));
1894 * Packet decoding: a packet is anything that the (ASF) demuxer contains,
1895 * and we expect that the demuxer / application provides it to us as such
1896 * (else you'll probably get garbage as output). Every packet has a size of
1897 * ctx->block_align bytes, starts with a packet header (see
1898 * #parse_packet_header()), and then a series of superframes. Superframe
1899 * boundaries may exceed packets, i.e. superframes can split data over
1900 * multiple (two) packets.
1902 * For more information about frames, see #synth_superframe().
1904 static int wmavoice_decode_packet(AVCodecContext *ctx, void *data,
1905 int *got_frame_ptr, AVPacket *avpkt)
1907 WMAVoiceContext *s = ctx->priv_data;
1908 GetBitContext *gb = &s->gb;
1911 /* Packets are sometimes a multiple of ctx->block_align, with a packet
1912 * header at each ctx->block_align bytes. However, FFmpeg's ASF demuxer
1913 * feeds us ASF packets, which may concatenate multiple "codec" packets
1914 * in a single "muxer" packet, so we artificially emulate that by
1915 * capping the packet size at ctx->block_align. */
1916 for (size = avpkt->size; size > ctx->block_align; size -= ctx->block_align);
1917 init_get_bits8(&s->gb, avpkt->data, size);
1919 /* size == ctx->block_align is used to indicate whether we are dealing with
1920 * a new packet or a packet of which we already read the packet header
1922 if (!(size % ctx->block_align)) { // new packet header
1924 s->spillover_nbits = 0;
1925 s->nb_superframes = 0;
1927 if ((res = parse_packet_header(s)) < 0)
1929 s->nb_superframes = res;
1932 /* If the packet header specifies a s->spillover_nbits, then we want
1933 * to push out all data of the previous packet (+ spillover) before
1934 * continuing to parse new superframes in the current packet. */
1935 if (s->sframe_cache_size > 0) {
1936 int cnt = get_bits_count(gb);
1937 if (cnt + s->spillover_nbits > avpkt->size * 8) {
1938 s->spillover_nbits = avpkt->size * 8 - cnt;
1940 copy_bits(&s->pb, avpkt->data, size, gb, s->spillover_nbits);
1941 flush_put_bits(&s->pb);
1942 s->sframe_cache_size += s->spillover_nbits;
1943 if ((res = synth_superframe(ctx, data, got_frame_ptr)) == 0 &&
1945 cnt += s->spillover_nbits;
1946 s->skip_bits_next = cnt & 7;
1950 skip_bits_long (gb, s->spillover_nbits - cnt +
1951 get_bits_count(gb)); // resync
1952 } else if (s->spillover_nbits) {
1953 skip_bits_long(gb, s->spillover_nbits); // resync
1955 } else if (s->skip_bits_next)
1956 skip_bits(gb, s->skip_bits_next);
1958 /* Try parsing superframes in current packet */
1959 s->sframe_cache_size = 0;
1960 s->skip_bits_next = 0;
1961 pos = get_bits_left(gb);
1962 if (s->nb_superframes-- == 0) {
1965 } else if (s->nb_superframes > 0) {
1966 if ((res = synth_superframe(ctx, data, got_frame_ptr)) < 0) {
1968 } else if (*got_frame_ptr) {
1969 int cnt = get_bits_count(gb);
1970 s->skip_bits_next = cnt & 7;
1974 } else if ((s->sframe_cache_size = pos) > 0) {
1975 /* ... cache it for spillover in next packet */
1976 init_put_bits(&s->pb, s->sframe_cache, SFRAME_CACHE_MAXSIZE);
1977 copy_bits(&s->pb, avpkt->data, size, gb, s->sframe_cache_size);
1978 // FIXME bad - just copy bytes as whole and add use the
1979 // skip_bits_next field
1985 static av_cold int wmavoice_decode_end(AVCodecContext *ctx)
1987 WMAVoiceContext *s = ctx->priv_data;
1990 ff_rdft_end(&s->rdft);
1991 ff_rdft_end(&s->irdft);
1992 ff_dct_end(&s->dct);
1993 ff_dct_end(&s->dst);
1999 AVCodec ff_wmavoice_decoder = {
2001 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio Voice"),
2002 .type = AVMEDIA_TYPE_AUDIO,
2003 .id = AV_CODEC_ID_WMAVOICE,
2004 .priv_data_size = sizeof(WMAVoiceContext),
2005 .init = wmavoice_decode_init,
2006 .close = wmavoice_decode_end,
2007 .decode = wmavoice_decode_packet,
2008 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
2009 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
2010 .flush = wmavoice_flush,