@scrypted/libav 1.0.166 → 1.0.168

package/opus/dnn/datasets.txt

@@ -0,0 +1,163 @@
+The following datasets can be used to train a language-independent FARGAN model
+and a Deep REDundancy (DRED) model. Note that this data typically needs to be
+resampled before it can be used.
+
+https://www.openslr.org/resources/30/si_lk.tar.gz
+https://www.openslr.org/resources/32/af_za.tar.gz
+https://www.openslr.org/resources/32/st_za.tar.gz
+https://www.openslr.org/resources/32/tn_za.tar.gz
+https://www.openslr.org/resources/32/xh_za.tar.gz
+https://www.openslr.org/resources/37/bn_bd.zip
+https://www.openslr.org/resources/37/bn_in.zip
+https://www.openslr.org/resources/41/jv_id_female.zip
+https://www.openslr.org/resources/41/jv_id_male.zip
+https://www.openslr.org/resources/42/km_kh_male.zip
+https://www.openslr.org/resources/43/ne_np_female.zip
+https://www.openslr.org/resources/44/su_id_female.zip
+https://www.openslr.org/resources/44/su_id_male.zip
+https://www.openslr.org/resources/61/es_ar_female.zip
+https://www.openslr.org/resources/61/es_ar_male.zip
+https://www.openslr.org/resources/63/ml_in_female.zip
+https://www.openslr.org/resources/63/ml_in_male.zip
+https://www.openslr.org/resources/64/mr_in_female.zip
+https://www.openslr.org/resources/65/ta_in_female.zip
+https://www.openslr.org/resources/65/ta_in_male.zip
+https://www.openslr.org/resources/66/te_in_female.zip
+https://www.openslr.org/resources/66/te_in_male.zip
+https://www.openslr.org/resources/69/ca_es_female.zip
+https://www.openslr.org/resources/69/ca_es_male.zip
+https://www.openslr.org/resources/70/en_ng_female.zip
+https://www.openslr.org/resources/70/en_ng_male.zip
+https://www.openslr.org/resources/71/es_cl_female.zip
+https://www.openslr.org/resources/71/es_cl_male.zip
+https://www.openslr.org/resources/72/es_co_female.zip
+https://www.openslr.org/resources/72/es_co_male.zip
+https://www.openslr.org/resources/73/es_pe_female.zip
+https://www.openslr.org/resources/73/es_pe_male.zip
+https://www.openslr.org/resources/74/es_pr_female.zip
+https://www.openslr.org/resources/75/es_ve_female.zip
+https://www.openslr.org/resources/75/es_ve_male.zip
+https://www.openslr.org/resources/76/eu_es_female.zip
+https://www.openslr.org/resources/76/eu_es_male.zip
+https://www.openslr.org/resources/77/gl_es_female.zip
+https://www.openslr.org/resources/77/gl_es_male.zip
+https://www.openslr.org/resources/78/gu_in_female.zip
+https://www.openslr.org/resources/78/gu_in_male.zip
+https://www.openslr.org/resources/79/kn_in_female.zip
+https://www.openslr.org/resources/79/kn_in_male.zip
+https://www.openslr.org/resources/80/my_mm_female.zip
+https://www.openslr.org/resources/83/irish_english_male.zip
+https://www.openslr.org/resources/83/midlands_english_female.zip
+https://www.openslr.org/resources/83/midlands_english_male.zip
+https://www.openslr.org/resources/83/northern_english_female.zip
+https://www.openslr.org/resources/83/northern_english_male.zip
+https://www.openslr.org/resources/83/scottish_english_female.zip
+https://www.openslr.org/resources/83/scottish_english_male.zip
+https://www.openslr.org/resources/83/southern_english_female.zip
+https://www.openslr.org/resources/83/southern_english_male.zip
+https://www.openslr.org/resources/83/welsh_english_female.zip
+https://www.openslr.org/resources/83/welsh_english_male.zip
+https://www.openslr.org/resources/86/yo_ng_female.zip
+https://www.openslr.org/resources/86/yo_ng_male.zip
+
+The corresponding citations for all these datasets are:
+
+  @inproceedings{demirsahin-etal-2020-open,
+    title = {{Open-source Multi-speaker Corpora of the English Accents in the British Isles}},
+    author = {Demirsahin, Isin and Kjartansson, Oddur and Gutkin, Alexander and Rivera, Clara},
+    booktitle = {Proceedings of The 12th Language Resources and Evaluation Conference (LREC)},
+    month = may,
+    year = {2020},
+    pages = {6532--6541},
+    address = {Marseille, France},
+    publisher = {European Language Resources Association (ELRA)},
+    url = {https://www.aclweb.org/anthology/2020.lrec-1.804},
+    ISBN = {979-10-95546-34-4},
+  }
+  @inproceedings{kjartansson-etal-2020-open,
+    title = {{Open-Source High Quality Speech Datasets for Basque, Catalan and Galician}},
+    author = {Kjartansson, Oddur and Gutkin, Alexander and Butryna, Alena and Demirsahin, Isin and Rivera, Clara},
+    booktitle = {Proceedings of the 1st Joint Workshop on Spoken Language Technologies for Under-resourced languages (SLTU) and Collaboration and Computing for Under-Resourced Languages (CCURL)},
+    year = {2020},
+    pages = {21--27},
+    month = may,
+    address = {Marseille, France},
+    publisher = {European Language Resources association (ELRA)},
+    url = {https://www.aclweb.org/anthology/2020.sltu-1.3},
+    ISBN = {979-10-95546-35-1},
+  }
+
+
+  @inproceedings{guevara-rukoz-etal-2020-crowdsourcing,
+    title = {{Crowdsourcing Latin American Spanish for Low-Resource Text-to-Speech}},
+    author = {Guevara-Rukoz, Adriana and Demirsahin, Isin and He, Fei and Chu, Shan-Hui Cathy and Sarin, Supheakmungkol and Pipatsrisawat, Knot and Gutkin, Alexander and Butryna, Alena and Kjartansson, Oddur},
+    booktitle = {Proceedings of The 12th Language Resources and Evaluation Conference (LREC)},
+    year = {2020},
+    month = may,
+    address = {Marseille, France},
+    publisher = {European Language Resources Association (ELRA)},
+    url = {https://www.aclweb.org/anthology/2020.lrec-1.801},
+    pages = {6504--6513},
+    ISBN = {979-10-95546-34-4},
+  }
+  @inproceedings{he-etal-2020-open,
+    title = {{Open-source Multi-speaker Speech Corpora for Building Gujarati, Kannada, Malayalam, Marathi, Tamil and Telugu Speech Synthesis Systems}},
+    author = {He, Fei and Chu, Shan-Hui Cathy and Kjartansson, Oddur and Rivera, Clara and Katanova, Anna and Gutkin, Alexander and Demirsahin, Isin and Johny, Cibu and Jansche, Martin and Sarin, Supheakmungkol and Pipatsrisawat, Knot},
+    booktitle = {Proceedings of The 12th Language Resources and Evaluation Conference (LREC)},
+    month = may,
+    year = {2020},
+    address = {Marseille, France},
+    publisher = {European Language Resources Association (ELRA)},
+    pages = {6494--6503},
+    url = {https://www.aclweb.org/anthology/2020.lrec-1.800},
+    ISBN = "{979-10-95546-34-4}",
+  }
+
+
+  @inproceedings{kjartansson-etal-tts-sltu2018,
+    title = {{A Step-by-Step Process for Building TTS Voices Using Open Source Data and Framework for Bangla, Javanese, Khmer, Nepali, Sinhala, and Sundanese}},
+    author = {Keshan Sodimana and Knot Pipatsrisawat and Linne Ha and Martin Jansche and Oddur Kjartansson and Pasindu De Silva and Supheakmungkol Sarin},
+    booktitle = {Proc. The 6th Intl. Workshop on Spoken Language Technologies for Under-Resourced Languages (SLTU)},
+    year  = {2018},
+    address = {Gurugram, India},
+    month = aug,
+    pages = {66--70},
+    URL   = {http://dx.doi.org/10.21437/SLTU.2018-14}
+  }
+
+
+  @inproceedings{oo-etal-2020-burmese,
+    title = {{Burmese Speech Corpus, Finite-State Text Normalization and Pronunciation Grammars with an Application to Text-to-Speech}},
+    author = {Oo, Yin May and Wattanavekin, Theeraphol and Li, Chenfang and De Silva, Pasindu and Sarin, Supheakmungkol and Pipatsrisawat, Knot and Jansche, Martin and Kjartansson, Oddur and Gutkin, Alexander},
+    booktitle = {Proceedings of The 12th Language Resources and Evaluation Conference (LREC)},
+    month = may,
+    year = {2020},
+    pages = "6328--6339",
+    address = {Marseille, France},
+    publisher = {European Language Resources Association (ELRA)},
+    url = {https://www.aclweb.org/anthology/2020.lrec-1.777},
+    ISBN = {979-10-95546-34-4},
+  }
+  @inproceedings{van-niekerk-etal-2017,
+    title = {{Rapid development of TTS corpora for four South African languages}},
+    author = {Daniel van Niekerk and Charl van Heerden and Marelie Davel and Neil Kleynhans and Oddur Kjartansson and Martin Jansche and Linne Ha},
+    booktitle = {Proc. Interspeech 2017},
+    pages = {2178--2182},
+    address = {Stockholm, Sweden},
+    month = aug,
+    year  = {2017},
+    URL   = {http://dx.doi.org/10.21437/Interspeech.2017-1139}
+  }
+
+  @inproceedings{gutkin-et-al-yoruba2020,
+    title = {{Developing an Open-Source Corpus of Yoruba Speech}},
+    author = {Alexander Gutkin and I{\c{s}}{\i}n Demir{\c{s}}ahin and Oddur Kjartansson and Clara Rivera and K\d{\'o}lá Túb\d{\`o}sún},
+    booktitle = {Proceedings of Interspeech 2020},
+    pages = {404--408},
+    month = {October},
+    year = {2020},
+    address = {Shanghai, China},
+    publisher = {International Speech and Communication Association (ISCA)},
+    doi = {10.21437/Interspeech.2020-1096},
+    url = {http://dx.doi.org/10.21437/Interspeech.2020-1096},
+  }

package/opus/dnn/download_model.bat

@@ -0,0 +1,9 @@
+@echo off
+set model=opus_data-%1.tar.gz
+
+if not exist %model% (
+    echo Downloading latest model
+    powershell -Command "(New-Object System.Net.WebClient).DownloadFile('https://media.xiph.org/opus/models/%model%', '%model%')"
+)
+
+tar -xvzf %model%

package/opus/dnn/download_model.sh

@@ -0,0 +1,30 @@
+#!/bin/sh
+set -e
+
+model=opus_data-$1.tar.gz
+
+if [ ! -f $model ]; then
+        echo "Downloading latest model"
+        wget https://media.xiph.org/opus/models/$model
+fi
+
+if command -v sha256sum
+then
+   echo "Validating checksum"
+   checksum="$1"
+   checksum2=$(sha256sum $model | awk '{print $1}')
+   if [ "$checksum" != "$checksum2" ]
+   then
+      echo "Aborting due to mismatching checksums. This could be caused by a corrupted download of $model."
+      echo "Consider deleting local copy of $model and running this script again."
+      exit 1
+   else
+      echo "checksums match"
+   fi
+else
+   echo "Could not find sha256 sum; skipping verification. Please verify manually that sha256 hash of ${model} matches ${1}."
+fi
+
+
+
+tar xvomf $model

package/opus/dnn/dred_coding.c

@@ -0,0 +1,44 @@
+/* Copyright (c) 2022 Amazon
+   Written by Jean-Marc Valin */
+/*
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions
+   are met:
+
+   - Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <math.h>
+
+#include "celt/entenc.h"
+#include "os_support.h"
+#include "dred_config.h"
+#include "dred_coding.h"
+
+int compute_quantizer(int q0, int dQ, int qmax, int i) {
+  int quant;
+  static const int dQ_table[8] = {0, 2, 3, 4, 6, 8, 12, 16};
+  quant = q0 + (dQ_table[dQ]*i + 8)/16;
+  return quant > qmax ? qmax : quant;
+}

package/opus/dnn/dred_coding.h

@@ -0,0 +1,36 @@
+/* Copyright (c) 2022 Amazon
+   Written by Jan Buethe */
+/*
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions
+   are met:
+
+   - Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef DRED_CODING_H
+#define DRED_CODING_H
+
+#include "opus_types.h"
+#include "entcode.h"
+
+int compute_quantizer(int q0, int dQ, int qmax, int i);
+
+#endif

package/opus/dnn/dred_config.h

@@ -0,0 +1,54 @@
+/* Copyright (c) 2022 Amazon
+   Written by Jan Buethe */
+/*
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions
+   are met:
+
+   - Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef DRED_CONFIG_H
+#define DRED_CONFIG_H
+
+/* Change this once DRED gets an extension number assigned. */
+#define DRED_EXTENSION_ID 126
+
+/* Remove these two completely once DRED gets an extension number assigned. */
+#define DRED_EXPERIMENTAL_VERSION 10
+#define DRED_EXPERIMENTAL_BYTES 2
+
+
+#define DRED_MIN_BYTES 8
+
+/* these are inpart duplicates to the values defined in dred_rdovae_constants.h */
+#define DRED_SILK_ENCODER_DELAY (79+12-80)
+#define DRED_FRAME_SIZE 160
+#define DRED_DFRAME_SIZE (2 * (DRED_FRAME_SIZE))
+#define DRED_MAX_DATA_SIZE 1000
+#define DRED_ENC_Q0 6
+#define DRED_ENC_Q1 15
+
+/* Covers 1.04 second so we can cover one second, after the lookahead. */
+#define DRED_MAX_LATENTS 26
+#define DRED_NUM_REDUNDANCY_FRAMES (2*DRED_MAX_LATENTS)
+#define DRED_MAX_FRAMES (4*DRED_MAX_LATENTS)
+
+#endif

package/opus/dnn/dred_decoder.c

@@ -0,0 +1,129 @@
+/* Copyright (c) 2022 Amazon
+   Written by Jan Buethe */
+/*
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions
+   are met:
+
+   - Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#include <string.h>
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include "os_support.h"
+#include "dred_decoder.h"
+#include "dred_coding.h"
+#include "celt/entdec.h"
+#include "celt/laplace.h"
+#include "dred_rdovae_stats_data.h"
+#include "dred_rdovae_constants.h"
+
+static void dred_decode_latents(ec_dec *dec, float *x, const opus_uint8 *scale, const opus_uint8 *r, const opus_uint8 *p0, int dim) {
+    int i;
+    for (i=0;i<dim;i++) {
+        int q;
+        if (r[i] == 0 || p0[i] == 255) q = 0;
+        else q = ec_laplace_decode_p0(dec, p0[i]<<7, r[i]<<7);
+        x[i] = q*256.f/(scale[i] == 0 ? 1 : scale[i]);
+    }
+}
+
+int dred_ec_decode(OpusDRED *dec, const opus_uint8 *bytes, int num_bytes, int min_feature_frames, int dred_frame_offset)
+{
+  ec_dec ec;
+  int q_level;
+  int i;
+  int offset;
+  int q0;
+  int dQ;
+  int qmax;
+  int state_qoffset;
+  int extra_offset;
+
+  /* since features are decoded in quadruples, it makes no sense to go with an uneven number of redundancy frames */
+  celt_assert(DRED_NUM_REDUNDANCY_FRAMES % 2 == 0);
+
+  /* decode initial state and initialize RDOVAE decoder */
+  ec_dec_init(&ec, (unsigned char*)bytes, num_bytes);
+  q0 = ec_dec_uint(&ec, 16);
+  dQ = ec_dec_uint(&ec, 8);
+  if (ec_dec_uint(&ec, 2)) extra_offset = 32*ec_dec_uint(&ec, 256);
+  else extra_offset = 0;
+  /* Compute total offset, including DRED position in a multiframe packet. */
+  dec->dred_offset = 16 - ec_dec_uint(&ec, 32) - extra_offset + dred_frame_offset;
+  /*printf("%d %d %d\n", dred_offset, q0, dQ);*/
+  qmax = 15;
+  if (q0 < 14 && dQ > 0) {
+    int nvals;
+    int ft;
+    int s;
+    /* The distribution for the dQmax symbol is split evenly between zero
+        (which implies qmax == 15) and larger values, with the probability of
+        all larger values being uniform.
+       This is equivalent to coding 1 bit to decide if the maximum is less than
+        15 followed by a uint to decide the actual value if it is less than
+        15, but combined into a single symbol. */
+    nvals = 15 - (q0 + 1);
+    ft = 2*nvals;
+    s = ec_decode(&ec, ft);
+    if (s >= nvals) {
+      qmax = q0 + (s - nvals) + 1;
+      ec_dec_update(&ec, s, s + 1, ft);
+    }
+    else {
+      ec_dec_update(&ec, 0, nvals, ft);
+    }
+  }
+  state_qoffset = q0*DRED_STATE_DIM;
+  dred_decode_latents(
+      &ec,
+      dec->state,
+      dred_state_quant_scales_q8 + state_qoffset,
+      dred_state_r_q8 + state_qoffset,
+      dred_state_p0_q8 + state_qoffset,
+      DRED_STATE_DIM);
+
+  /* decode newest to oldest and store oldest to newest */
+  for (i = 0; i < IMIN(DRED_NUM_REDUNDANCY_FRAMES, (min_feature_frames+1)/2); i += 2)
+  {
+      /* FIXME: Figure out how to avoid missing a last frame that would take up < 8 bits. */
+      if (8*num_bytes - ec_tell(&ec) <= 7)
+         break;
+      q_level = compute_quantizer(q0, dQ, qmax, i/2);
+      offset = q_level*DRED_LATENT_DIM;
+      dred_decode_latents(
+          &ec,
+          &dec->latents[(i/2)*DRED_LATENT_DIM],
+          dred_latent_quant_scales_q8 + offset,
+          dred_latent_r_q8 + offset,
+          dred_latent_p0_q8 + offset,
+          DRED_LATENT_DIM
+          );
+
+      offset = 2 * i * DRED_NUM_FEATURES;
+  }
+  dec->process_stage = 1;
+  dec->nb_latents = i/2;
+  return i/2;
+}

package/opus/dnn/dred_decoder.h

@@ -0,0 +1,49 @@
+/* Copyright (c) 2022 Amazon
+   Written by Jan Buethe */
+/*
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions
+   are met:
+
+   - Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef DRED_DECODER_H
+#define DRED_DECODER_H
+
+#include "opus.h"
+#include "dred_config.h"
+#include "dred_rdovae.h"
+#include "entcode.h"
+#include "dred_rdovae_constants.h"
+
+struct OpusDRED {
+    float        fec_features[2*DRED_NUM_REDUNDANCY_FRAMES*DRED_NUM_FEATURES];
+    float        state[DRED_STATE_DIM];
+    float        latents[(DRED_NUM_REDUNDANCY_FRAMES/2)*DRED_LATENT_DIM];
+    int          nb_latents;
+    int          process_stage;
+    int          dred_offset;
+};
+
+
+int dred_ec_decode(OpusDRED *dec, const opus_uint8 *bytes, int num_bytes, int min_feature_frames, int dred_frame_offset);
+
+#endif