- Add the take_over_pcm preference, persistence and input registration.
- Decode takeover sources with ffmpeg to 48 kHz float32 and feed the kernel
PCM_LAYOUT stream for binaural output.
- Keep JOC priority: bare E-AC-3/EC3 and MP4/Matroska/WebM are probed first,
then non-JOC audio falls back to ffmpeg.
- Cover AC3, DTS/DTS-HD, WavPack, APE, TTA, MPC, AU/SND/CAF and TrueHD/MLP;
TrueHD in containers follows the same fallback.
- Sync the kernel to 1.3.3 with the virtual_speaker module and fix the PCM
gain path.
- Update the preferences UI, HRTF hint and README.
The kernel copy follows JustOneCacophonyCXX 1.3.1: a frame without JOC metadata
mid-stream now holds the previous matrices instead of switching to the core
planes, which removes the clicks the 6.0 kernels produced on encoders that drop
the JOC metadata for single frames.
resolve_bed_channels() called ReadFile once and scanned that single chunk. A
read returns only the bytes that are available (a pipe hands over a partial
chunk), and an encoder may leave the first audio frames without JOC metadata,
so the first JOC frame could sit past the window. The scan then fell back to 6
bed channels, and the first JOC frame (joc_dmx_config_idx=1, 8 channels) was
rejected with core_channel_count_mismatch. The scan now reads 64 KiB chunks
until the first syncframe carrying the JOC EMDF container is seen, keeps the
trailing partial frame between reads, and rewinds the stream as before; the
container path and the bare-file path both use it.
Bump the component version to 0.6.1, and pin the kernel notices to 1.3.1.
Verified with the offline harness: the reported GNR M4A now reports
"joc_dmx_config_idx=1 -> 8 bed channel(s)" and decodes the whole track (8906
frames), and the Smooth Criminal opening no longer steps at the two frames
without metadata (6.016 s and 7.328 s).
The kernel copy follows JustOneCacophonyCXX abce846 (version 1.3.0). An ID14
payload that ends inside joc_data is now a frame status (7) instead of a
stream error, so the renderer holds the previous matrices and playback
continues; the file task counts those frames and the stream status struct
gains an appended truncated_joc_frames field that joc_stream_status()
accepts in both struct sizes. The nine kernel files are byte-identical to
the upstream checkout again, and the component needs no code change because
the fix lives inside the vendored kernel.
README, THIRD_PARTY_NOTICES, the kernel project and the CI comment now pin
the copy to abce846 (version 1.3.0).
Built Release-Static|x64. The offline test tools build and pass, and
render_harness decodes the reported M4A to the end: 8 channels, 13,441,536
samples (8751 frames), peak 0.742114365, byte-identical to the fixed core
CLI output.
The kernel copy follows the new kernel. The renderer has a frame-level state
machine (normal, hold, hard cut and passthrough, the 8-plane passthrough, the
LFE clipgain ramp and interpolation when no previous state exists), the OAMD
parser carries the complete frame state (several blocks, more than 16 objects)
and the speaker renderer applies the metadata (gain and inactive flags,
zone/elevation, size, snap, divergence, ISF and bed distribution). oamd_math and
oamd_params are new modules, the ISF tables are compiled in now that the
upstream data/ directory is gone, and the binaural timeline accepts a complete
frame state. joc_core.h moves to ABI 6 (frame parameters and EMDF payload info
at v2, JOC_FRAME_STATUS_*, raw 5-bit offset_ts) and eac3joc_core.h to EJOC ABI 4
(EJOC_FRAME_STATUS_*, ejoc_frame_input v1, ejoc_renderer_render_frame,
EJOC_CORE_OUT_CHANNELS and EJOC_LFE_SLOT; the old ejoc_renderer_process stays).
joc_stream.h and joc_version.h are unchanged.
joc_kernel.vcxproj lists the two new translation units and the four new
headers. The component's layout table gains the kernel's 6.0 layout (L R Ls Rs
Lrs Rrs) after 9.1.6, and layout_table_test expects twelve layouts, so the list
the preferences page shows and the list the renderer accepts are the same
again. README, THIRD_PARTY_NOTICES and the CI comment pin the copy to commit
3f0d5e87fe41b90d9350528d2dca2ca82a637a3f; tag v1.2.0 points at the older
a276028 and is not the sync baseline.
Builds Release-Static|x64 and |Win32 with no errors. The offline test tools
build and pass on both architectures: layout_table_test reports 12 layouts, and
channel_swap_test, log_rotation_test and long_path_test pass. scan_crosscheck
agrees with the core on the upstream test material, and the render harness
delivers 7.1.4 speaker output and binaural output on both architectures, with
the speaker-path WAV byte-identical between them.
The core is still vendored under kernel/; building it from the upstream
checkout at build time is a separate change.
The kernel copy follows the new kernel: the OAMD parser reads the trim
element, the speaker renderer applies warp, per-object trim and the
front-back balance, and the ADM writer carries them into the BWF metadata.
joc_kernel.vcxproj lists the new speaker_trim translation unit and header,
and the copied joc_version.h reports kernel 1.2.0 in the session log.
The component keeps its playback settings, its speaker layout table and
the side/rear channel swap; the stream interface is unchanged, so the
decoder wrapper needs no edit. A stream whose OAMD asks for warp 2/3 now
stops at the speaker renderer with JOC_ERR_NOT_SUPPORTED, as the reference
refuses it.
Builds Release-Static|x64 and |Win32. layout_table_test and channel_swap_test
still pass; the render harness renders a warp-1 stream and reports the
unimplemented mode for a warp-3 stream.
The kernel copy follows the new kernel: the SOFA reader, its compile cache and
the field fit are gone, the compiled-field renderer is jochrtf_renderer.cpp with
jochrtf_runtime.*, and joc_kernel.vcxproj lists the new translation units.
With the Rosella model as the only user-facing input, the preferences page loses
its SOFA radio and the settings drop HrtfSource, hrtf_source,
hrtf_cache_policy, hrtf_cache_dir and hrtf_radius_m. Decoding always sets
personalized_headphone_path, the default lookup is
HRTF\binaural.personalized_headphone, and the page hint names that file alone.
The compiled .jochrtf stays internal: the page names a model file, never a cache.
Builds Release-Static|x64 (foo_input_joc.dll and joc_kernel.lib).
Bring the kernel copy up to the upstream state that adds the captured
Rosella filterbank and the bank selector: bank 1 is the shared public
bank (13 taps / 6 slots, 961 samples), bank 2 is the captured bank
(11 taps / 3 slots, 448 samples). The copy was still at the state
before the filterbank fix, so both upstream changes land here.
- Add src/hrtf/rosella_bank.{h,cpp}, rosella_filterbank.{h,cpp} and
rosella_filterbank_tables.h, and list them in joc_kernel.vcxproj
- Carry the updated joc_core.h / joc_stream.h, the renderer, the
stream and task paths, and the kernel CLI's --rosella-filterbank
The host code needs no change: joc_task_config and joc_stream_config
keep their sizes (the reserved word became rosella_filterbank, and the
task config version went 3 -> 4), so src/joc_decode.cpp builds as is.
Checked: joc_kernel.vcxproj builds for Release-Static|x64, and
`diff -rq kernel\src <upstream>\src` plus the include tree are clean.