* Windows Media Audio Voice decoder.
* Copyright (c) 2009 Ronald S. Bultje
*
- * This file is part of FFmpeg.
+ * This file is part of Libav.
*
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
- * FFmpeg is distributed in the hope that it will be useful,
+ * Libav is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
+ * License along with Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
* @author Ronald S. Bultje <rsbultje@gmail.com>
*/
+#define UNCHECKED_BITSTREAM_READER 1
+
#include <math.h>
+
+#include "libavutil/channel_layout.h"
+#include "libavutil/float_dsp.h"
+#include "libavutil/mem.h"
#include "avcodec.h"
+#include "internal.h"
#include "get_bits.h"
#include "put_bits.h"
#include "wmavoice_data.h"
-#include "celp_math.h"
#include "celp_filters.h"
#include "acelp_vectors.h"
#include "acelp_filters.h"
#include "lsp.h"
-#include "libavutil/lzo.h"
-#include "avfft.h"
-#include "fft.h"
+#include "dct.h"
+#include "rdft.h"
+#include "sinewin.h"
#define MAX_BLOCKS 8 ///< maximum number of blocks per frame
#define MAX_LSPS 16 ///< maximum filter order
*/
typedef struct {
/**
- * @defgroup struct_global Global values
- * Global values, specified in the stream header / extradata or used
- * all over.
+ * @name Global values specified in the stream header / extradata or used all over.
* @{
*/
GetBitContext gb; ///< packet bitreader. During decoder init,
/**
* @}
- * @defgroup struct_packet Packet values
- * Packet values, specified in the packet header or related to a packet.
+ *
+ * @name Packet values specified in the packet header or related to a packet.
+ *
* A packet is considered to be a single unit of data provided to this
* decoder by the demuxer.
* @{
*/
int spillover_nbits; ///< number of bits of the previous packet's
- ///< last superframe preceeding this
+ ///< last superframe preceding this
///< packet's first full superframe (useful
///< for re-synchronization also)
int has_residual_lsps; ///< if set, superframes contain one set of
/**
* @}
- * @defgroup struct_frame Frame and superframe values
+ *
+ * @name Frame and superframe values
* Superframe and frame data - these can change from frame to frame,
* although some of them do in that case serve as a cache / history for
* the next frame or superframe.
float synth_history[MAX_LSPS]; ///< see #excitation_history
/**
* @}
- * @defgroup post_filter Postfilter values
- * Varibales used for postfilter implementation, mostly history for
+ *
+ * @name Postfilter values
+ *
+ * Variables used for postfilter implementation, mostly history for
* smoothing and so on, and context variables for FFT/iFFT.
* @{
*/
///< by postfilter
float denoise_filter_cache[MAX_FRAMESIZE];
int denoise_filter_cache_size; ///< samples in #denoise_filter_cache
- DECLARE_ALIGNED(16, float, tilted_lpcs_pf)[0x80];
+ DECLARE_ALIGNED(32, float, tilted_lpcs_pf)[0x80];
///< aligned buffer for LPC tilting
- DECLARE_ALIGNED(16, float, denoise_coeffs_pf)[0x80];
+ DECLARE_ALIGNED(32, float, denoise_coeffs_pf)[0x80];
///< aligned buffer for denoise coefficients
- DECLARE_ALIGNED(16, float, synth_filter_out_buf)[0x80 + MAX_LSPS_ALIGN16];
+ DECLARE_ALIGNED(32, float, synth_filter_out_buf)[0x80 + MAX_LSPS_ALIGN16];
///< aligned buffer for postfilter speech
///< synthesis
/**
0x0ffc, 0x0ffd, 0x0ffe, // 1111111111+00/01/10
0x3ffc, 0x3ffd, 0x3ffe, 0x3fff // 111111111111+xx
};
- int cntr[8], n, res;
+ int cntr[8] = { 0 }, n, res;
- memset(vbm_tree, 0xff, sizeof(vbm_tree));
- memset(cntr, 0, sizeof(cntr));
+ memset(vbm_tree, 0xff, sizeof(vbm_tree[0]) * 25);
for (n = 0; n < 17; n++) {
res = get_bits(gb, 3);
if (cntr[res] > 3) // should be >= 3 + (res == 7))
s->min_pitch_val = ((ctx->sample_rate << 8) / 400 + 50) >> 8;
s->max_pitch_val = ((ctx->sample_rate << 8) * 37 / 2000 + 50) >> 8;
pitch_range = s->max_pitch_val - s->min_pitch_val;
+ if (pitch_range <= 0) {
+ av_log(ctx, AV_LOG_ERROR, "Invalid pitch range; broken extradata?\n");
+ return -1;
+ }
s->pitch_nbits = av_ceil_log2(pitch_range);
s->last_pitch_val = 40;
s->last_acb_type = ACB_TYPE_NONE;
s->block_conv_table[2] = (pitch_range * 44) >> 6;
s->block_conv_table[3] = s->max_pitch_val - 1;
s->block_delta_pitch_hrange = (pitch_range >> 3) & ~0xF;
+ if (s->block_delta_pitch_hrange <= 0) {
+ av_log(ctx, AV_LOG_ERROR, "Invalid delta pitch hrange; broken extradata?\n");
+ return -1;
+ }
s->block_delta_pitch_nbits = 1 + av_ceil_log2(s->block_delta_pitch_hrange);
s->block_pitch_range = s->block_conv_table[2] +
s->block_conv_table[3] + 1 +
2 * (s->block_conv_table[1] - 2 * s->min_pitch_val);
s->block_pitch_nbits = av_ceil_log2(s->block_pitch_range);
- ctx->sample_fmt = SAMPLE_FMT_FLT;
+ ctx->channels = 1;
+ ctx->channel_layout = AV_CH_LAYOUT_MONO;
+ ctx->sample_fmt = AV_SAMPLE_FMT_FLT;
return 0;
}
/**
- * @defgroup postfilter Postfilter functions
+ * @name Postfilter functions
* Postfilter functions (gain control, wiener denoise filter, DC filter,
* kalman smoothening, plus surrounding code to wrap it)
* @{
/* find best fitting point in history */
do {
- dot = ff_dot_productf(in, ptr, size);
+ dot = avpriv_scalarproduct_float_c(in, ptr, size);
if (dot > optimal_gain) {
optimal_gain = dot;
best_hist_ptr = ptr;
if (optimal_gain <= 0)
return -1;
- dot = ff_dot_productf(best_hist_ptr, best_hist_ptr, size);
+ dot = avpriv_scalarproduct_float_c(best_hist_ptr, best_hist_ptr, size);
if (dot <= 0) // would be 1.0
return -1;
{
float rh0, rh1;
- rh0 = 1.0 + ff_dot_productf(lpcs, lpcs, n_lpcs);
- rh1 = lpcs[0] + ff_dot_productf(lpcs, &lpcs[1], n_lpcs - 1);
+ rh0 = 1.0 + avpriv_scalarproduct_float_c(lpcs, lpcs, n_lpcs);
+ rh1 = lpcs[0] + avpriv_scalarproduct_float_c(lpcs, &lpcs[1], n_lpcs - 1);
return rh1 / rh0;
}
int n, idx;
/* Create frequency power spectrum of speech input (i.e. RDFT of LPCs) */
- ff_rdft_calc(&s->rdft, lpcs);
+ s->rdft.rdft_calc(&s->rdft, lpcs);
#define log_range(var, assign) do { \
float tmp = log10f(assign); var = tmp; \
max = FFMAX(max, tmp); min = FFMIN(min, tmp); \
* is a sinus input) by doing a phase shift (in theory, H(sin())=cos()).
* Hilbert_Transform(RDFT(x)) = Laplace_Transform(x), which calculates the
* "moment" of the LPCs in this filter. */
- ff_dct_calc(&s->dct, lpcs);
- ff_dct_calc(&s->dst, lpcs);
+ s->dct.dct_calc(&s->dct, lpcs);
+ s->dst.dct_calc(&s->dst, lpcs);
/* Split out the coefficient indexes into phase/magnitude pairs */
idx = 255 + av_clip(lpcs[64], -255, 255);
coeffs[1] = last_coeff;
/* move into real domain */
- ff_rdft_calc(&s->irdft, coeffs);
+ s->irdft.rdft_calc(&s->irdft, coeffs);
/* tilt correction and normalize scale */
memset(&coeffs[remainder], 0, sizeof(coeffs[0]) * (128 - remainder));
-1.8 * tilt_factor(coeffs, remainder - 1),
coeffs, remainder);
}
- sq = (1.0 / 64.0) * sqrtf(1 / ff_dot_productf(coeffs, coeffs, remainder));
+ sq = (1.0 / 64.0) * sqrtf(1 / avpriv_scalarproduct_float_c(coeffs, coeffs,
+ remainder));
for (n = 0; n < remainder; n++)
coeffs[n] *= sq;
}
/* apply coefficients (in frequency spectrum domain), i.e. complex
* number multiplication */
memset(&synth_pf[size], 0, sizeof(synth_pf[0]) * (128 - size));
- ff_rdft_calc(&s->rdft, synth_pf);
- ff_rdft_calc(&s->rdft, coeffs);
+ s->rdft.rdft_calc(&s->rdft, synth_pf);
+ s->rdft.rdft_calc(&s->rdft, coeffs);
synth_pf[0] *= coeffs[0];
synth_pf[1] *= coeffs[1];
for (n = 1; n < 64; n++) {
synth_pf[n * 2] = v1 * coeffs[n * 2] - v2 * coeffs[n * 2 + 1];
synth_pf[n * 2 + 1] = v2 * coeffs[n * 2] + v1 * coeffs[n * 2 + 1];
}
- ff_rdft_calc(&s->irdft, synth_pf);
+ s->irdft.rdft_calc(&s->irdft, synth_pf);
}
/* merge filter output with the history of previous runs */
}
/**
- * @defgroup lsp_dequant LSP dequantization routines
+ * @name LSP dequantization routines
* LSP dequantization routines, for 10/16LSPs and independent/residual coding.
* @note we assume enough bits are available, caller should check.
* lsp10i() consumes 24 bits; lsp10r() consumes an additional 24 bits;
/**
* @}
- * @defgroup aw Pitch-adaptive window coding functions
+ * @name Pitch-adaptive window coding functions
* The next few functions are for pitch-adaptive window coding.
* @{
*/
int excl_range = s->aw_pulse_range; // always 16 or 24
uint16_t *use_mask_ptr = &use_mask[idx >> 4];
int first_sh = 16 - (idx & 15);
- *use_mask_ptr++ &= 0xFFFF << first_sh;
+ *use_mask_ptr++ &= 0xFFFFu << first_sh;
excl_range -= first_sh;
if (excl_range >= 16) {
*use_mask_ptr++ = 0;
/* Calculate gain for adaptive & fixed codebook signal.
* see ff_amr_set_fixed_gain(). */
idx = get_bits(gb, 7);
- fcb_gain = expf(ff_dot_productf(s->gain_pred_err, gain_coeff, 6) -
+ fcb_gain = expf(avpriv_scalarproduct_float_c(s->gain_pred_err,
+ gain_coeff, 6) -
5.2409161640 + wmavoice_gain_codebook_fcb[idx]);
acb_gain = wmavoice_gain_codebook_acb[idx];
pred_err = av_clipf(wmavoice_gain_codebook_fcb[idx],
int pitch[MAX_BLOCKS], last_block_pitch;
/* Parse frame type ("frame header"), see frame_descs */
- int bd_idx = s->vbm_tree[get_vlc2(gb, frame_type_vlc.table, 6, 3)],
- block_nsamples = MAX_FRAMESIZE / frame_descs[bd_idx].n_blocks;
+ int bd_idx = s->vbm_tree[get_vlc2(gb, frame_type_vlc.table, 6, 3)], block_nsamples;
if (bd_idx < 0) {
av_log(ctx, AV_LOG_ERROR,
return -1;
}
+ block_nsamples = MAX_FRAMESIZE / frame_descs[bd_idx].n_blocks;
+
/* Pitch calculation for ACB_TYPE_ASYMMETRIC ("pitch-per-frame") */
if (frame_descs[bd_idx].acb_type == ACB_TYPE_ASYMMETRIC) {
/* Pitch is provided per frame, which is interpreted as the pitch of
* (if less than 480), usually used to prevent blanks at track boundaries.
*
* @param ctx WMA Voice decoder context
- * @param samples pointer to output buffer for voice samples
- * @param data_size pointer containing the size of #samples on input, and the
- * amount of #samples filled on output
* @return 0 on success, <0 on error or 1 if there was not enough data to
* fully parse the superframe
*/
-static int synth_superframe(AVCodecContext *ctx,
- float *samples, int *data_size)
+static int synth_superframe(AVCodecContext *ctx, AVFrame *frame,
+ int *got_frame_ptr)
{
WMAVoiceContext *s = ctx->priv_data;
GetBitContext *gb = &s->gb, s_gb;
wmavoice_mean_lsf16[s->lsp_def_mode] : wmavoice_mean_lsf10[s->lsp_def_mode];
float excitation[MAX_SIGNAL_HISTORY + MAX_SFRAMESIZE + 12];
float synth[MAX_LSPS + MAX_SFRAMESIZE];
+ float *samples;
memcpy(synth, s->synth_history,
s->lsps * sizeof(*synth));
s->sframe_cache_size = 0;
}
- if ((res = check_bits_for_superframe(gb, s)) == 1) return 1;
+ if ((res = check_bits_for_superframe(gb, s)) == 1) {
+ *got_frame_ptr = 0;
+ return 1;
+ }
/* First bit is speech/music bit, it differentiates between WMAVoice
* speech samples (the actual codec) and WMAVoice music samples, which
* are really WMAPro-in-WMAVoice-superframes. I've never seen those in
* the wild yet. */
if (!get_bits1(gb)) {
- av_log_missing_feature(ctx, "WMAPro-in-WMAVoice support", 1);
- return -1;
+ avpriv_request_sample(ctx, "WMAPro-in-WMAVoice");
+ return AVERROR_PATCHWELCOME;
}
/* (optional) nr. of samples in superframe; always <= 480 and >= 0 */
stabilize_lsps(lsps[n], s->lsps);
}
- /* Parse frames, optionally preceeded by per-frame (independent) LSPs. */
+ /* get output buffer */
+ frame->nb_samples = 480;
+ if ((res = ff_get_buffer(ctx, frame, 0)) < 0) {
+ av_log(ctx, AV_LOG_ERROR, "get_buffer() failed\n");
+ return res;
+ }
+ frame->nb_samples = n_samples;
+ samples = (float *)frame->data[0];
+
+ /* Parse frames, optionally preceded by per-frame (independent) LSPs. */
for (n = 0; n < 3; n++) {
if (!s->has_residual_lsps) {
int m;
&samples[n * MAX_FRAMESIZE],
lsps[n], n == 0 ? s->prev_lsps : lsps[n - 1],
&excitation[s->history_nsamples + n * MAX_FRAMESIZE],
- &synth[s->lsps + n * MAX_FRAMESIZE])))
+ &synth[s->lsps + n * MAX_FRAMESIZE]))) {
+ *got_frame_ptr = 0;
return res;
+ }
}
/* Statistics? FIXME - we don't check for length, a slight overrun
skip_bits(gb, 10 * (res + 1));
}
- /* Specify nr. of output samples */
- *data_size = n_samples * sizeof(float);
+ *got_frame_ptr = 1;
/* Update history */
memcpy(s->prev_lsps, lsps[2],
* @param size size of the source data, in bytes
* @param gb bit I/O context specifying the current position in the source.
* data. This function might use this to align the bit position to
- * a whole-byte boundary before calling #ff_copy_bits() on aligned
+ * a whole-byte boundary before calling #avpriv_copy_bits() on aligned
* source data
* @param nbits the amount of bits to copy from source to target
*
rmn_bits = rmn_bytes = get_bits_left(gb);
if (rmn_bits < nbits)
return;
+ if (nbits > pb->size_in_bits - put_bits_count(pb))
+ return;
rmn_bits &= 7; rmn_bytes >>= 3;
if ((rmn_bits = FFMIN(rmn_bits, nbits)) > 0)
put_bits(pb, rmn_bits, get_bits(gb, rmn_bits));
- ff_copy_bits(pb, data + size - rmn_bytes,
+ avpriv_copy_bits(pb, data + size - rmn_bytes,
FFMIN(nbits - rmn_bits, rmn_bytes << 3));
}
* For more information about frames, see #synth_superframe().
*/
static int wmavoice_decode_packet(AVCodecContext *ctx, void *data,
- int *data_size, AVPacket *avpkt)
+ int *got_frame_ptr, AVPacket *avpkt)
{
WMAVoiceContext *s = ctx->priv_data;
GetBitContext *gb = &s->gb;
int size, res, pos;
- if (*data_size < 480 * sizeof(float)) {
- av_log(ctx, AV_LOG_ERROR,
- "Output buffer too small (%d given - %zu needed)\n",
- *data_size, 480 * sizeof(float));
- return -1;
- }
- *data_size = 0;
-
/* Packets are sometimes a multiple of ctx->block_align, with a packet
- * header at each ctx->block_align bytes. However, FFmpeg's ASF demuxer
+ * header at each ctx->block_align bytes. However, Libav's ASF demuxer
* feeds us ASF packets, which may concatenate multiple "codec" packets
* in a single "muxer" packet, so we artificially emulate that by
* capping the packet size at ctx->block_align. */
for (size = avpkt->size; size > ctx->block_align; size -= ctx->block_align);
- if (!size)
+ if (!size) {
+ *got_frame_ptr = 0;
return 0;
+ }
init_get_bits(&s->gb, avpkt->data, size << 3);
/* size == ctx->block_align is used to indicate whether we are dealing with
copy_bits(&s->pb, avpkt->data, size, gb, s->spillover_nbits);
flush_put_bits(&s->pb);
s->sframe_cache_size += s->spillover_nbits;
- if ((res = synth_superframe(ctx, data, data_size)) == 0 &&
- *data_size > 0) {
+ if ((res = synth_superframe(ctx, data, got_frame_ptr)) == 0 &&
+ *got_frame_ptr) {
cnt += s->spillover_nbits;
s->skip_bits_next = cnt & 7;
return cnt >> 3;
s->sframe_cache_size = 0;
s->skip_bits_next = 0;
pos = get_bits_left(gb);
- if ((res = synth_superframe(ctx, data, data_size)) < 0) {
+ if ((res = synth_superframe(ctx, data, got_frame_ptr)) < 0) {
return res;
- } else if (*data_size > 0) {
+ } else if (*got_frame_ptr) {
int cnt = get_bits_count(gb);
s->skip_bits_next = cnt & 7;
return cnt >> 3;
}
}
-AVCodec wmavoice_decoder = {
- "wmavoice",
- AVMEDIA_TYPE_AUDIO,
- CODEC_ID_WMAVOICE,
- sizeof(WMAVoiceContext),
- wmavoice_decode_init,
- NULL,
- wmavoice_decode_end,
- wmavoice_decode_packet,
- CODEC_CAP_SUBFRAMES,
- .flush = wmavoice_flush,
- .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio Voice"),
+AVCodec ff_wmavoice_decoder = {
+ .name = "wmavoice",
+ .type = AVMEDIA_TYPE_AUDIO,
+ .id = AV_CODEC_ID_WMAVOICE,
+ .priv_data_size = sizeof(WMAVoiceContext),
+ .init = wmavoice_decode_init,
+ .close = wmavoice_decode_end,
+ .decode = wmavoice_decode_packet,
+ .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
+ .flush = wmavoice_flush,
+ .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio Voice"),
};