Files
foo_input_joc/README.en.md
T
TheM14 5c0f9ea2ff
build / windows (push) Canceled after 0s
build / release (push) Canceled after 0s
Sync the 1.3.1 kernel and read the stream until its first JOC frame
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).
2026-11-10 15:35:32 +08:00

213 lines
13 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# foo_input_joc — foobar2000 input component for E-AC-3 JOC
[简体中文](README.md) · [Install](#install) · [Settings](#settings) · [Building](#building) · [Known limitations](#known-limitations)
A foobar2000 input component for **E-AC-3 JOC (Dolby Atmos)** files: the JOC object and OAMD
metadata is taken from the E-AC-3 syncframes and paired with the core PCM that ffmpeg decodes
(5.1 or 7 channels, following $joc\_dmx\_config\_idx$), then rendered in real time to headphones
(HRTF) or to a speaker layout from 2.0 up to 9.1.6. The
rendering core is compiled into the component, so there is nothing to install beside
`foo_input_joc.dll`.
Input is a bare `.eac3` / `.ec3` stream, or an E-AC-3 JOC track inside a container (`.mp4`
`.m4a` `.m4b` `.m4p` `.m4r` `.mov` `.mkv` `.mka` `.webm`). A file without JOC is not claimed:
plain AC-3 / E-AC-3, a container whose audio track is AAC, and transport streams are all handed
back to foobar2000 and play through the decoder it would have used anyway.
## Requirements
* foobar2000 1.6 (32-bit) or 2.x (32-bit and 64-bit).
* An `ffmpeg` executable, found on `PATH` by default; the preferences page can name one.
* For binaural output, an HRTF data file. **This repository does not ship one** — see
[HRTF data](#hrtf-data).
## Install
Download the package for your architecture from [Releases](../../releases) — a pushed `v*` tag
makes CI attach both — or build it yourself as described under [Building](#building) and take the
result from `dist\`; either way the file is `foo_input_joc-<version>-<arch>.fb2k-component`. Drop
it onto foobar2000, or use Preferences → Components → Install, and restart. Take `-x86` for
foobar2000 1.6 and 2.x 32-bit, `-x64` for 2.x 64-bit.
To install by hand, copy `foo_input_joc.dll` into the per-component subdirectory of the
`user-components` folder the running version reads:
| foobar2000 | Folder |
|---|---|
| 1.6 | `<profile>\user-components\foo_input_joc\` |
| 2.x | `<app>\user-components\foo_input_joc\` (also in portable mode) |
The subdirectory is required — a DLL lying directly in `user-components\` is not scanned — and a
component placed in the folder the running version does not read, or built for the other
architecture, is **ignored without a message**.
`tools/deploy.ps1 -TestBed <portable foobar2000>` is the scripted form of the manual steps, and
`tools/run.ps1 -TestBed <path> -Play <file>` plays a file unattended and prints the log. Let
foobar2000 exit through `/exit`: a force-killed instance leaves a `<profile>\running` marker
behind, and the next start then refuses to load any user component.
### Containers need one look at the decoder list
foobar2000 asks the decoders in the order shown in Preferences → **Decoding**, and the built-in
container readers are in that list. When one of them is offered an MP4 or Matroska file first, it
takes the file and the JOC objects are lost — such a file then plays as plain E-AC-3.
So move **JOC decoder (E-AC-3 JOC)** above **foobar2000 MP4 Demuxer** and **foobar2000
Matroska/WebM Reader** in that list. Bare `.eac3` / `.ec3` files do not depend on the order.
## Settings
Preferences → Tools → **JOC decoder**:
* **Output** — **Binaural (HRTF)**, or a **Speaker layout** from 2.0 to 9.1.6 (names and channel order
in the [table below](#speaker-layouts-and-channel-order));
* **Binaural mode** (near / mid / far) and the room **tail** in seconds;
* **HRTF** — a **Rosella personalized model**. An empty path means the default
location, `<component directory>\HRTF\binaural.personalized_headphone`;
* **Gain** — a switch and a value in dB. Binaural rendering can exceed full scale on material
that does not clip in the core mix, so attenuation belongs here;
* **Force 24-bit output (int24)** — off by default, which delivers float32. When on, every sample
is quantised with the core's own int24 rule (clip to ±1, ×8388607, truncate) before it is handed
over, the samples that had to be clipped are counted in the log, and the component reports the
file as 24-bit fixed point. The renderer itself stays float32: this is the precision the decoder
delivers, not the output device's bit depth, which is set in Preferences → Output. **Reload a
file's information after flipping this switch** (Properties → *Reload info*, or remove the file
from the Media Library and add it again): foobar2000 caches technical information per file, and
until it is reloaded Properties and the Converter still go by the old 32-bit floating-point
answer, warning that the encoder does not take floating-point input.
* **Swap side and rear channels** — off by default. When on, `Ls` and `Lrs`, and
`Rs` and `Rrs`, are exchanged just before the PCM is handed over (5↔7 and 6↔8
counting from one in 7.1). It only applies to layouts that carry both a side and
a rear pair — 7.1, 7.1.2, 7.1.4, 9.1.4 and 9.1.6 — so the top channels in the
last two slots of 5.1.2 and 5.1.4 are never touched. Only the delivered channel
order changes; decoding, object positions and the layout table are untouched,
and both the float32 and the int24 delivery carry the swap.
* the **ffmpeg.exe** executable to use.
* **Language** — **Follow system** / **简体中文** / **English**. Every string on the page, the status
lines and the errors a user can see use the selected language, and switching takes effect
immediately; the choice is stored in the component's own configuration. "Follow system" goes by
the Windows user-interface language: Chinese (including Traditional) reads the simplified Chinese
column, everything else reads English. The setting only affects interface text; decoding and
rendering never read it.
No control on the page is disabled, and the status line states what is in effect. The control names
above are the English ones; the Chinese page names the same controls 输出, 扬声器布局, 双耳模式,
HRTF 文件, 增益.
### HRTF data
A SOFA measurement set or a personalised headphone model is supplied by whoever runs the
component, and is listed in `.gitignore` so it cannot be committed by accident. Speaker layouts
need none. Binaural rendering without an HRTF fails with a message naming the file it looked for.
## Speaker layouts and channel order
The PCM this component hands to foobar2000, and the multichannel WAV files the Converter writes,
follow the renderer's channel order:
| Layout | Channel order |
|---|---|
| 2.0 | L, R |
| 3.1 | L, R, C, LFE |
| 4.0 | L, R, Lrs, Rrs |
| 5.1 | L, R, C, LFE, Ls, Rs |
| 7.1 | L, R, C, LFE, Ls, Rs, Lrs, Rrs |
| 5.1.2 | L, R, C, LFE, Ls, Rs, Ltm, Rtm |
| 5.1.4 | L, R, C, LFE, Ls, Rs, Ltf, Rtf, Ltr, Rtr |
| 7.1.2 | L, R, C, LFE, Ls, Rs, Lrs, Rrs, Ltm, Rtm |
| 7.1.4 | L, R, C, LFE, Ls, Rs, Lrs, Rrs, Ltf, Rtf, Ltr, Rtr |
| 9.1.4 | L, R, C, LFE, Ls, Rs, Lrs, Rrs, Lw, Rw, Ltf, Rtf, Ltr, Rtr |
| 9.1.6 | L, R, C, LFE, Ls, Rs, Lrs, Rrs, Lw, Rw, Ltf, Rtf, Ltm, Rtm, Ltr, Rtr |
`4.0` is the quad layout this project adds (the traditional `FL FR BL BR`, here `L R Lrs Rrs`): a
front and a back pair with neither a centre nor an LFE — an object straight ahead is a phantom image
of the front pair, and the LFE in the input has no output channel in this layout, so by the core's
rule it is not mixed into the others. Every other layout and channel order follows the Dolby
specifications.
**Channel order and the foobar2000 limitation.** The component hands the renderer's PCM over
unchanged, in the order of the table above; when that stream is written to WAV (the playback chunk or
a Converter output), the order of the 1–16 channels follows the Dolby documentation exactly. The
channel *definitions*, however, are not guaranteed to correspond fully: foobar2000's WAVE writer can
only declare a Windows speaker mask, which has no slot for top-middle (`Ts`) or wide (`Ww`) and cannot
express an order such as 7.1's `Ls Rs Lrs Rrs`. This is a foobar2000 limitation, not a problem with
the rendered order. Properties declares the channel count and no speaker positions; the chunk used
while playing carries foobar2000's default map for that channel count (an unspecified map is rejected
as an invalid audio stream specification and playback fails).
For a multichannel WAV with `dwChannelMask = 0`, download the CLI from
[github.com/TheM14/JustOneCacophony](https://github.com/TheM14/JustOneCacophony) and export with it: it
writes `WAVE_FORMAT_EXTENSIBLE` with `dwChannelMask = 0` and the order above.
Mapping the channels to a player, ASIO, a DAW or a device is up to the user; the component guarantees
the content and numbering of each PCM channel as listed above.
Layout names in the configuration follow the table above; a name that cannot be resolved is logged
and falls back to `7.1`.
## Building
```powershell
pwsh -File tools/setup_sdk.ps1 # official SDK into SDK/, pinned to target 1.5/1.6
pwsh -File tools/build.ps1 # Win32 -> build\Win32\foo_input_joc.dll
pwsh -File tools/build.ps1 -Platform x64
pwsh -File tools/package.ps1 # both, packaged into dist\*.fb2k-component
```
The configuration is fixed at `Release-Static` (static CRT, `/MT`), and `/fp:precise` is what the
byte-for-byte acceptance rests on, so it must not be changed. `foo_input_joc.vcxproj` builds
`kernel\joc_kernel.vcxproj` first through a project reference; the rendering core sources in
`kernel/` are compiled with `JOC_STATIC` / `EJOC_STATIC`, so their entry points are neither
imported nor exported. `tests\` holds the offline tools (bitstream self-test and cross-check,
render comparison, speaker-layout table and channel-order check, preferences-page layout check,
Chinese and English text-width check, container-probe check) and `tools\` the build and test-bed
scripts. Settings also read `JOC_*`
environment overrides for one run (development only); the list and what each one does is in
`src\settings.cpp`.
To diagnose a problem, read `joc_decoder.log` beside the DLL: the component writes its own
version, the core version and the log path there at start-up. The log is capped by **bytes**: a
file that reaches 16 MiB is rolled to `joc_decoder.log.1` beside it, replacing the previous roll
rather than accumulating, so the two together never exceed 32 MiB and the newest lines are always
in `joc_decoder.log`. The version banner is written again after a roll. A Media Library scan logs
every container in the library, which is where the volume comes from.
## Known limitations
* ADM BWF output is not implemented.
* Speaker layouts go up to 9.1.6 (16 channels), the renderer's own output limit; see
[Speaker layouts and channel order](#speaker-layouts-and-channel-order).
* A multichannel layout declares no speaker positions in Properties, and the per-chunk map used for
playback is only foobar2000's default for the channel count; output mapping is up to the user's
channel assignment (same section).
* All five $joc\_dmx\_config\_idx$ configurations are supported: the bed is paired per audio
frame, and its channel count follows the configuration (6 channels for 0/3, 8 for 1/2/4), with
`-ac 6` passed to ffmpeg only for the 6-channel ones. An AC-3 core plus a dependent substream
(`strmtyp=1`) and a first audio frame without JOC are both normal cases.
* A container is only claimed when this component is ahead of the built-in container reader in
Preferences → Decoding (see [Install](#containers-need-one-look-at-the-decoder-list)); the core
does not let a decoder ask for a file another entry has already taken. Such a file's tags stay
that reader's as well.
* Tags of a bare `.eac3` / `.ec3` (an ID3v2 tag in front of the stream, or an APEv2/ID3v1 tag
behind it) can be **read but not written**: no component claims raw E-AC-3 for writing, and
rewriting the whole file to insert a tag is not this component's job.
* Transport streams (`.ts`, `.m2ts`) are not claimed.
* **Force 24-bit output** only changes the precision the decoder delivers: foobar2000 converts the
int24 it receives into its own floating-point samples, and which bit depth the output device runs
at is still decided in Preferences → Output, not by this component. Samples outside ±1 follow the
core's `continue` clip policy (binaural rendering can exceed full scale, so use **Gain** when
attenuation is wanted).
* Playback length is exactly the file's duration. The binaural renderer still computes its room
tail, but it is not delivered as playback time the file does not have.
* A seek re-enters the bitstream at the frame holding the target instead of decoding everything
in front of it — which is why the cost of a seek does not depend on where it lands. The
position is exact and does not drift; the samples are the same waveform handed to a decoder
that started there, so they differ from a straight play-through by a low-level noise floor of
−59 dBFS or quieter.
## Licence
`LICENSE` is the upstream MIT licence, copied unchanged; `kernel/` is a copy of the upstream
rendering core sources, synced to JustOneCacophonyCXX 1.3.1, and keeps their notices. See
[THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md).