metasound: add last missing modes (8kHz @ 6kbps per channel)

Signed-off-by: Anton Khirnov <[email protected]>

Changelog

@@ -32,7 +32,7 @@ version 10:
  the -noaccurate_seek option.
 - avconv -t option can now be used for inputs, to limit the duration of
  data read from an input file
-- incomplete Voxware MetaSound decoder
+- Voxware MetaSound decoder
 - WebP decoder
 - Error Resilient AAC syntax (ER AAC LC) decoding
 - Low Delay AAC (ER AAC LD) decoding

doc/general.texi

@@ -827,7 +827,6 @@ following image formats are supported:
 @item Vorbis @tab E @tab X
  @tab A native but very primitive encoder exists.
 @item Voxware MetaSound @tab @tab X
- @tab imperfect and incomplete support
 @item WavPack @tab E @tab X
  @tab supported through external library libwavpack
 @item Westwood Audio (SND1) @tab @tab X

libavcodec/metasound.c

@@ -187,7 +187,7 @@ static int metasound_read_bitstream(AVCodecContext *avctx, TwinVQContext *tctx,
 
  sub = mtab->fmode[bits->ftype].sub;
 
- if (bits->ftype != TWINVQ_FT_SHORT)
+ if (bits->ftype != TWINVQ_FT_SHORT && !tctx->is_6kbps)
  get_bits(&gb, 2);
 
  read_cb_data(tctx, &gb, bits->main_coeffs, bits->ftype);
@@ -307,6 +307,12 @@ static av_cold int metasound_decode_init(AVCodecContext *avctx)
  ibps = avctx->bit_rate / (1000 * avctx->channels);
 
  switch ((avctx->channels << 16) + (isampf << 8) + ibps) {
+ case (1 << 16) + ( 8 << 8) + 6:
+ tctx->mtab = &ff_metasound_mode0806;
+ break;
+ case (2 << 16) + ( 8 << 8) + 6:
+ tctx->mtab = &ff_metasound_mode0806s;
+ break;
  case (1 << 16) + ( 8 << 8) + 8:
  tctx->mtab = &ff_metasound_mode0808;
  break;
@@ -362,6 +368,7 @@ static av_cold int metasound_decode_init(AVCodecContext *avctx)
  tctx->decode_ppc = decode_ppc;
  tctx->frame_size = avctx->bit_rate * tctx->mtab->size
  / avctx->sample_rate;
+ tctx->is_6kbps = ibps == 6;
 
  return ff_twinvq_decode_init(avctx);
 }

libavcodec/metasound_data.c

@@ -15190,6 +15190,24 @@ static const uint16_t bark_tab_s44_128[] = {
  1, 2, 1, 2, 3, 4, 6, 10, 23, 76
 };
 
+const TwinVQModeTab ff_metasound_mode0806 = {
+ {
+ { 8, bark_tab_s8_64, 10, fcb8s, 1, 5, cb0806ss0, cb0806ss1, 27 },
+ { 2, bark_tab_m8_256, 20, fcb8m, 2, 5, cb0806sm0, cb0806sm1, 22 },
+ { 1, bark_tab_l8_512, 30, fcb8l, 3, 6, cb0806sl0, cb0806sl1, 24 }
+ },
+ 512, 12, lsp8, 1, 5, 3, 3, shape8, 8, 28, 20, 6, 200
+};
+
+const TwinVQModeTab ff_metasound_mode0806s = {
+ {
+ { 8, bark_tab_s8s_64, 10, fcb8ss, 1, 5, cb0806ss0, cb0806ss1, 27 },
+ { 2, bark_tab_m8s_256, 20, fcb8sm, 2, 5, cb0806sm0, cb0806sm1, 22 },
+ { 1, bark_tab_l8s_512, 30, fcb8sl, 3, 6, cb0806sl0, cb0806sl1, 24 }
+ },
+ 512, 12, lsp8s, 1, 5, 3, 3, shape8s, 8, 28, 20, 6, 200
+};
+
 const TwinVQModeTab ff_metasound_mode0808 = {
  {
  { 8, bark_tab_s8_64, 10, fcb8s, 1, 5, cb0808s0, cb0808s1, 18 },

libavcodec/twinvq.c

@@ -706,7 +706,7 @@ static av_cold void init_bitstream_params(TwinVQContext *tctx)
  TWINVQ_WINDOW_TYPE_BITS +
  mtab->fmode[i].sub * (bse_bits[i] + n_ch * TWINVQ_SUB_GAIN_BITS);
 
- if (tctx->codec == TWINVQ_CODEC_METASOUND) {
+ if (tctx->codec == TWINVQ_CODEC_METASOUND && !tctx->is_6kbps) {
  bsize_no_main_cb[1] += 2;
  bsize_no_main_cb[2] += 2;
  }

libavcodec/twinvq.h

@@ -141,6 +141,8 @@ typedef struct TwinVQContext {
 
  const TwinVQModeTab *mtab;
 
+ int is_6kbps;
+
  // history
  float lsp_hist[2][20]; ///< LSP coefficients of the last frame
  float bark_hist[3][2][40]; ///< BSE coefficients of last frame

libavcodec/twinvqdec.c

@@ -402,6 +402,7 @@ static av_cold int twinvq_decode_init(AVCodecContext *avctx)
  tctx->decode_ppc = decode_ppc;
  tctx->frame_size = avctx->bit_rate * tctx->mtab->size
  / avctx->sample_rate + 8;
+ tctx->is_6kbps = 0;
  if (avctx->block_align && avctx->block_align * 8 / tctx->frame_size > 1) {
  av_log(avctx, AV_LOG_ERROR,
  "VQF TwinVQ should have only one frame per packet\n");