AAC: LOAS/LATM: defer parsing LATM frames until first AudioSpecificConfig is found

fixes #3727

NEWS.md

@@ -1,5 +1,12 @@
 # Version ?
 
+## Bug fixes
+
+* mkvmerge: AAC parser: LOAS/LATM streams: if the first `AudioSpecificConfig`
+  element is not found within the first LOAS/LATM frame, `mkvmerge` will no
+  longer discard the frames before it but defer parsing them until after it
+  has found the `AudioSpecificConfig` element. Fixes #3727.
+
 ## Build system changes
 
 * The bundled `fmt` library was updated to v10.2.1.

src/common/aac.cpp

@@ -710,32 +710,96 @@ parser_c::decode_loas_latm_header(uint8_t const *buffer,
 std::pair<parser_c::parse_result_e, size_t>
 parser_c::decode_header(uint8_t const *buffer,
                         size_t buffer_size) {
-  if (adif_multiplex == m_multiplex_type)
-    return { failure, 0 };
-
   if (adts_multiplex == m_multiplex_type)
     return decode_adts_header(buffer, buffer_size);
 
   if (loas_latm_multiplex == m_multiplex_type)
     return decode_loas_latm_header(buffer, buffer_size);
 
-  auto result = decode_adts_header(buffer, buffer_size);
+  return { failure, 0 };
+}
+
+bool
+parser_c::determine_multiplex_type(uint8_t const *new_buffer,
+                                   std::size_t new_buffer_size) {
+  at_scope_exit_c cleanup{[this]() {
+    m_copy_data = true;
+
+    if (m_multiplex_type != unknown_multiplex)
+      m_multiplex_type_detection_buffer.clear();
+  }};
+
+  m_multiplex_type_detection_buffer.add(new_buffer, new_buffer_size);
+
+  m_copy_data      = false;
+
+  auto buffer      = m_multiplex_type_detection_buffer.get_buffer();
+  auto buffer_size = m_multiplex_type_detection_buffer.get_size();
+  auto result      = decode_adts_header(buffer, buffer_size);
+
   if (result.first == success) {
+    mxdebug_if(m_debug, fmt::format("determine_multiplex_type: successfully detected ADTS at the start\n"));
     m_multiplex_type = adts_multiplex;
-    return result;
+    return true;
   }
 
   result = decode_loas_latm_header(buffer, buffer_size);
+
   if (result.first == success) {
+    mxdebug_if(m_debug, fmt::format("determine_multiplex_type: successfully detected LOAS LATM at the start\n"));
     m_multiplex_type = loas_latm_multiplex;
-    return result;
+    return true;
   }
 
-  return result;
+  std::vector<std::pair<multiplex_type_e, unsigned int>> multiplexes_to_try{ { loas_latm_multiplex, 0u }, { adif_multiplex, 0u } };
+
+  for (auto &multiplex : multiplexes_to_try) {
+    auto position = 0u;
+
+    while (position < buffer_size) {
+      auto remaining_bytes = buffer_size - position;
+      result               = multiplex.first == loas_latm_multiplex ? decode_loas_latm_header(&buffer[position], remaining_bytes) : decode_adts_header(&buffer[position], remaining_bytes);
+
+      if (result.first == need_more_data)
+        break;
+
+      auto num_bytes = std::max<std::size_t>(std::min(result.second, remaining_bytes), 1);
+      position      += num_bytes;
+
+      if (result.first == success)
+        ++multiplex.second;
+    }
+  }
+
+  auto latm_config_parsed = m_latm_parser.config_parsed();
+
+  std::optional<unsigned int> winner;
+
+  for (unsigned int idx = 0; idx < multiplexes_to_try.size(); ++idx) {
+    mxdebug_if(m_debug, fmt::format("determine_multiplex_type: idx {0} type {1} size {2}\n", idx, static_cast<unsigned int>(multiplexes_to_try[idx].first), multiplexes_to_try[idx].second));
+
+    auto num_frames = multiplexes_to_try[idx].second;
+
+    if ((num_frames > 0) && (!winner || (num_frames > multiplexes_to_try[*winner].second)))
+      winner = idx;
+  }
+
+  if (winner.has_value()) {
+    m_multiplex_type = multiplexes_to_try[*winner].first;
+    mxdebug_if(m_debug, fmt::format("determine_multiplex_type: we have a winner: {0} LATM configuration parsed: {1}\n", static_cast<unsigned int>(m_multiplex_type), latm_config_parsed));
+    return true;
+  }
+
+  mxdebug_if(m_debug, fmt::format("determine_multiplex_type: no winner\n"));
+
+  return false;
 }
 
 void
 parser_c::push_frame(frame_c &frame) {
+  if (m_multiplex_type == unknown_multiplex)
+    return;
+
   if (!m_provided_timestamps.empty()) {
     frame.m_timestamp = m_provided_timestamps.front();
     m_provided_timestamps.pop_front();
@@ -763,6 +827,9 @@ parser_c::parse() {
   auto buffer_size = m_fixed_buffer ? m_fixed_buffer_size : m_buffer.get_size();
   auto position    = 0u;
 
+  if ((m_multiplex_type == unknown_multiplex) && !determine_multiplex_type(buffer, buffer_size))
+    return;
+
   while (position < buffer_size) {
     auto remaining_bytes = buffer_size - position;
     auto result          = decode_header(&buffer[position], remaining_bytes);

src/common/aac.h

@@ -169,7 +169,7 @@ protected:
 protected:
   std::deque<frame_c> m_frames;
   std::deque<timestamp_c> m_provided_timestamps;
-  mtx::bytes::buffer_c m_buffer;
+  mtx::bytes::buffer_c m_buffer, m_multiplex_type_detection_buffer;
   uint8_t const *m_fixed_buffer;
   size_t m_fixed_buffer_size;
   uint64_t m_parsed_stream_position, m_total_stream_position;
@@ -216,6 +216,7 @@ protected:
   std::pair<parse_result_e, size_t> decode_header(uint8_t const *buffer, size_t buffer_size);
   std::pair<parse_result_e, size_t> decode_adts_header(uint8_t const *buffer, size_t buffer_size);
   std::pair<parse_result_e, size_t> decode_loas_latm_header(uint8_t const *buffer, size_t buffer_size);
+  bool determine_multiplex_type(uint8_t const *buffer, std::size_t buffer_size);
   void push_frame(frame_c &frame);
 };
 using parser_cptr = std::shared_ptr<parser_c>;

tests/results.txt

@@ -611,3 +611,4 @@ T_0763vp9_alpha_channel_data:ac329c9d810d9b19fb3ffcde6f12af2a-OK:passed:20231203
 T_0764ui_locale_be_BY:a44c54eadfb4c8fbdc104b75aa1de1c1-72b98d331b58a0f95e10159fca191b52:passed:20240120-191944:0.043782405
 T_0765ffmpeg_metadata_chapters:f16630c4019413c98b75b959a5697391-6b2b843310e80367b5fe5aaa8a5d51c4:passed:20240310-145016:0.047790171
 T_0766ui_locale_nb_NO:6e0054bcf8d381306adc9d4d212d1f6a-5a0be94aab291615f8ebd47f887e6eba:passed:20240422-215240:0.044197325
+T_0767aac_latm_audio_specific_config_late:6ef4bed121ea8c3bcb94d9e460d5c67a:passed:20240730-160739:0.022664154

tests/test-0767aac_latm_audio_specific_config_late.rb

@@ -0,0 +1,5 @@
+#!/usr/bin/ruby -w
+
+# T_767aac_latm_audio_specific_config_late
+describe "mkvmerge / AAC LOAS/LATM stream, AudioSpecificConfig comes in later frame"
+test_merge "data/aac/aac_lc_latm_problem.aac"