Apply the OAMD trim and front-back balance in speaker rendering

Select the trim configuration that applies to the output layout from its
surround and height speaker counts:

    cfg = kappa_f(surrounds) + 3 * kappa_h(heights)

Trim and balance apply to seven layouts: 2.0, 3.1, 4.0, 5.1, 5.1.2, 5.1.4
and 7.1.  Wider layouts are left untrimmed.

An explicit configuration contributes a static level for the centre,
surround and height groups, which multiplies the panning gains per output
channel.  Automatic and disabled configurations contribute no static
level: the attenuation an automatic configuration describes is already
applied per object by the layout position compensation.

An explicit configuration also contributes the two front-back balance
values.  Writing n = round(16 |b|) and s = n / 18, each balance
reparameterises the OAMD front-back axis in [0, 1]:

    y <= 0.05 or y >= 0.95        -> y
    y < 0.5,  b > 0               -> (1 + s) y - 0.05 s
    y >= 0.5, b > 0               -> (1 - s) (y - 0.5) + 0.5 + 0.45 s
    y < 0.5,  b < 0               -> (1 - s) y + 0.05 s
    y >= 0.5, b < 0               -> (1 + s) (y - 0.5) + 0.5 - 0.45 s

The map is monotonic onto [0.05, 0.95], keeps both walls fixed, and needs
no clamping.  Height blends the listener-plane and overhead values before
panning; the horizontal and vertical coordinates are untouched.

Warp is not read here: it selects a downmix mode and does not enter the
speaker path.
This commit is contained in:
TheM14 committed 2026-11-02 17:18:53 +08:00
1 parent 220e791ce9
commit 8424e6bdfb
6 files changed
+420 -12

No files matched your search

+117 -3
View File
@@ -519,7 +519,119 @@ G_{\mathrm{pos}}
\mathbf G_{\mathrm{point}}. \mathbf G_{\mathrm{point}}.
$$ $$
## 12. OAMD time alignment and gain ramps ## 12. OAMD trim and front-back balance
### 12.1 Configuration selection
The OAMD trim element splits a programme's calibration data into nine
configurations, chosen by the target layout's additional horizontal speaker
count $N_f$ and height speaker count $N_h$:
$$
\kappa_f(N)=
\begin{cases}
0, & N=0,\\
1, & N\in\{2,3\},\\
2, & \text{otherwise},
\end{cases}
\qquad
\kappa_h(N)=
\begin{cases}
0, & N=0,\\
1, & N=2,\\
2, & \text{otherwise}.
\end{cases}
$$
$$
c_{\mathrm{trim}}=\kappa_f(N_f)+3\,\kappa_h(N_h).
$$
The nine configurations enumerate "no / some / many additional horizontal
speakers by no / some / many height speakers".
Trim and front-back balance apply to seven layouts only: 2.0, 3.1, 4.0, 5.1,
5.1.2, 5.1.4 and 7.1. Wider layouts (7.1.2, 7.1.4, 9.1.4, 9.1.6) are left
untrimmed, with $G_c=1$ and $b=0$.
### 12.2 Group levels
Each configuration supplies one level $T_g$ in dB for the centre, surround and
height groups. Only an explicit configuration contributes a static level:
$$
T_g=
\begin{cases}
\text{coded value}, & \text{explicit},\
0, & \text{disabled},\
0, & \text{automatic}.
\end{cases}
$$
An automatic configuration carries no static group gain: the attenuation it
describes is already applied per object by the position compensation of §11,
and multiplying it in a second time would count it twice.
The static gain of channel $c$ is
$$
G_c=10^{T_{g(c)}/20},
$$
where $g(c)$ follows the channel label: centre maps to centre, side, rear and
wide map to surround, top maps to height, and left, right and LFE stay at
$G_c=1$.
### 12.3 Front-back balance
Each configuration also carries two front-back balance values:
$b_{\mathrm{lis}}$ for the listener plane and $b_{\mathrm{tb}}$ for the overhead
layer. Both are quantised to a $1/16$ grid and are non-zero only in an explicit
configuration.
Writing $n=\mathrm{round}(16\,|b|)$ and $s=n/18$, the single-layer map is
$$
B_b(y)=
\begin{cases}
y, & y\le 0.05\ \text{or}\ y\ge 0.95,\\[2pt]
(1+s)y-0.05s, & 0.05<y<0.5,\ b>0,\\[2pt]
(1-s)(y-0.5)+0.5+0.45s, & 0.5\le y<0.95,\ b>0,\\[2pt]
(1-s)y+0.05s, & 0.05<y<0.5,\ b<0,\\[2pt]
(1+s)(y-0.5)+0.5-0.45s, & 0.5\le y<0.95,\ b<0.
\end{cases}
$$
It maps $[0.05,0.95]$ monotonically onto itself: the two walls stay fixed, the
two halves meet at $y=0.5$, and their slopes are $1+s$ and $1-s$. The 5 percent
wall bands and the case $b=0$ pass through unchanged, so the map never leaves
the interval and needs no extra clamp.
The height $w$ blends the two layers:
$$
y'=(1-w)\,B_{b_{\mathrm{lis}}}(y)+w\,B_{b_{\mathrm{tb}}}(y).
$$
The front-back axis $y$ is oriented as in §9, with $0$ at the front wall and
$1$ at the back wall; $x$ and $z$ are unchanged.
### 12.4 Where it is applied
The balance acts on the second component of the object position before
panning:
$$
y_o\longmapsto y_o',
$$
while the group gain multiplies the panning result, that is $g_{o,c}$ in the
mix formula of §14 becomes $G_c\,g_{o,c}$.
Warp takes no part in either step: it selects a downmix mode, affects only the
downmix path, and leaves object coordinates untouched.
## 13. OAMD time alignment and gain ramps
The coded position of an OAMD update is The coded position of an OAMD update is
@@ -575,7 +687,7 @@ $$
If no new metadata update intervenes, this is equivalent to a sample-wise linear ramp of total length $KB$. If no new metadata update intervenes, this is equivalent to a sample-wise linear ramp of total length $KB$.
## 13. Final speaker mix ## 14. Final speaker mix
For target output channel $c$: For target output channel $c$:
@@ -607,9 +719,11 @@ y_{24}[n]=
\right). \right).
$$ $$
## 14. Scope of the formulas ## 15. Scope of the formulas
- The JOC matrix section covers both the dense MTX and the sparse IDX/VEC differential syntax. - The JOC matrix section covers both the dense MTX and the sparse IDX/VEC differential syntax.
- The speaker-panning section describes ordinary point objects; extent, spread, divergence, and similar modes require additional models. - The speaker-panning section describes ordinary point objects; extent, spread, divergence, and similar modes require additional models.
- Multiple OAMD position blocks must be scheduled in time order. - Multiple OAMD position blocks must be scheduled in time order.
- A limiter is separate post-processing and is not included in the mixing equations above. - A limiter is separate post-processing and is not included in the mixing equations above.
- Trim and front-back balance apply to speaker rendering only;
binaural rendering reads neither of them.
+104 -3
View File
@@ -516,7 +516,107 @@ G_{\mathrm{pos}}
\mathbf G_{\mathrm{point}}. \mathbf G_{\mathrm{point}}.
$$ $$
## 12. OAMD 时间对齐与增益斜坡 ## 12. OAMD trim 与前后平衡
### 12.1 配置选择
OAMD trim element 把节目的校准数据分成九个配置,由目标布局的附加水平扬声器数
$N_f$ 与高度扬声器数 $N_h$ 选定:
$$
\kappa_f(N)=
\begin{cases}
0, & N=0,\\
1, & N\in\{2,3\},\\
2, & \text{其他},
\end{cases}
\qquad
\kappa_h(N)=
\begin{cases}
0, & N=0,\\
1, & N=2,\\
2, & \text{其他}.
\end{cases}
$$
$$
c_{\mathrm{trim}}=\kappa_f(N_f)+3\,\kappa_h(N_h).
$$
九个配置依次为“no / some / many 附加水平 × no / some / many 高度”。
trim 与前后平衡只在七种布局下施加:2.0、3.1、4.0、5.1、5.1.2、5.1.4、7.1。
更宽的布局(7.1.2、7.1.4、9.1.4、9.1.6)不施加,取 $G_c=1$、$b=0$。
### 12.2 组电平
每个配置给 centre、surround、height 三个组各一个电平 $T_g$(dB)。只有显式配置
施加静态电平:
$$
T_g=
\begin{cases}
\text{记录值}, & \text{显式},\
0, & \text{禁用},\
0, & \text{自动}.
\end{cases}
$$
自动配置不带静态组增益:它描述的衰减量已由 §11 的位置补偿按对象逐个施加,再乘
一次会重复计算。
声道 $c$ 的静态增益为
$$
G_c=10^{T_{g(c)}/20},
$$
$g(c)$ 由声道标签决定:中置归 centre,侧、后、宽归 surround,顶部归 height,左右
与 LFE 不参与,取 $G_c=1$。
### 12.3 前后平衡
每个配置还携带两个前后平衡量:$b_{\mathrm{lis}}$ 作用于听音平面,
$b_{\mathrm{tb}}$ 作用于顶上层层。二者按 $1/16$ 网格量化,只在显式配置下非零。
记 $n=\mathrm{round}(16\,|b|)$、$s=n/18$,单层映射为
$$
B_b(y)=
\begin{cases}
y, & y\le 0.05\ \text{或}\ y\ge 0.95,\\[2pt]
(1+s)y-0.05s, & 0.05<y<0.5,\ b>0,\\[2pt]
(1-s)(y-0.5)+0.5+0.45s, & 0.5\le y<0.95,\ b>0,\\[2pt]
(1-s)y+0.05s, & 0.05<y<0.5,\ b<0,\\[2pt]
(1+s)(y-0.5)+0.5-0.45s, & 0.5\le y<0.95,\ b<0.
\end{cases}
$$
它把 $[0.05,0.95]$ 单调映到自身:两端不动,$y=0.5$ 处两段取值相同,斜率分别为
$1+s$ 与 $1-s$。首尾各 $5\%$ 的墙区以及 $b=0$ 的情形都原样通过,因此映射不会
越界,也无需额外钳位。
高度 $w$ 把两层混合起来:
$$
y'=(1-w)\,B_{b_{\mathrm{lis}}}(y)+w\,B_{b_{\mathrm{tb}}}(y).
$$
前后轴 $y$ 的取向见 §9,$0$ 为前、$1$ 为后;$x$ 与 $z$ 不变。
### 12.4 施加位置
前后平衡在声像之前作用于对象坐标的第二个分量:
$$
y_o\longmapsto y_o',
$$
组增益则乘在声像结果上,即 §14 混音式中的 $g_{o,c}$ 替换为 $G_c\,g_{o,c}$。
warp 不参与上述任何一步:它是下混模式的选择位,只影响下混路径,不改变对象坐标。
## 13. OAMD 时间对齐与增益斜坡
OAMD 更新的编码位置为 OAMD 更新的编码位置为
@@ -572,7 +672,7 @@ $$
如果中途没有新的 metadata 更新,该过程等价于总长度 $KB$ 的逐样本线性斜坡。 如果中途没有新的 metadata 更新,该过程等价于总长度 $KB$ 的逐样本线性斜坡。
## 13. 最终扬声器混音 ## 14. 最终扬声器混音
对目标输出声道 $c$: 对目标输出声道 $c$:
@@ -604,9 +704,10 @@ y_{24}[n]=
\right). \right).
$$ $$
## 14. 公式适用范围 ## 15. 公式适用范围
- JOC 矩阵部分同时描述 dense MTX 与 sparse IDX/VEC 两条差分语法。 - JOC 矩阵部分同时描述 dense MTX 与 sparse IDX/VEC 两条差分语法。
- 扬声器声像部分描述普通点对象;extent、spread、divergence 等模式需要额外模型。 - 扬声器声像部分描述普通点对象;extent、spread、divergence 等模式需要额外模型。
- 多个 OAMD position block 必须按其时间顺序调度。 - 多个 OAMD position block 必须按其时间顺序调度。
- limiter 属于独立后处理,不包含在上述混音公式中。 - limiter 属于独立后处理,不包含在上述混音公式中。
- trim 与前后平衡只作用于扬声器渲染;双耳渲染不读取它们。
+3 -1
View File
@@ -699,8 +699,10 @@ def main(argv=None):
timings["create_binaural_renderer"] = 0.0 timings["create_binaural_renderer"] = 0.0
layout = get_speaker_layout(args.speaker_layout) layout = get_speaker_layout(args.speaker_layout)
speaker_name = speaker_layout_display_name(layout) speaker_name = speaker_layout_display_name(layout)
speaker_meta = oamd_tracks.read_atmos_meta(index)
speaker_decoder, speaker_backend_info = create_speaker_renderer( speaker_decoder, speaker_backend_info = create_speaker_renderer(
layout, bed_channels=getattr(index, "bed_channels", ())) layout, bed_channels=getattr(index, "bed_channels", ()),
trim_meta=speaker_meta)
print(f"[speaker] layout={speaker_name} backend={speaker_backend_info['name']} " print(f"[speaker] layout={speaker_name} backend={speaker_backend_info['name']} "
f"channels={layout.channel_count}", flush=True) f"channels={layout.channel_count}", flush=True)
spool = SpeakerPcmSpool( spool = SpeakerPcmSpool(
+4 -2
View File
@@ -11,10 +11,12 @@ from speaker_renderer import PythonSpeakerRenderer, payload_events_from_index, r
FRAME_SAMPLES = 1536 FRAME_SAMPLES = 1536
def create_speaker_renderer(layout: str | SpeakerLayout, bed_channels=None): def create_speaker_renderer(layout: str | SpeakerLayout, bed_channels=None,
trim_meta=None):
"""Create a stateful speaker renderer and return ``(renderer, info)``.""" """Create a stateful speaker renderer and return ``(renderer, info)``."""
target = get_speaker_layout(layout) if isinstance(layout, str) else layout target = get_speaker_layout(layout) if isinstance(layout, str) else layout
renderer = PythonSpeakerRenderer(target, bed_channels=bed_channels) renderer = PythonSpeakerRenderer(target, bed_channels=bed_channels,
trim_meta=trim_meta)
return renderer, { return renderer, {
"name": "python", "name": "python",
"layout": target.name, "layout": target.name,
+11 -3
View File
@@ -14,6 +14,7 @@ import numpy as np
from oamd_bits import JocFieldState, frame_update from oamd_bits import JocFieldState, frame_update
from speaker_layouts import RegionGeometry, SpeakerLayout, get_speaker_layout from speaker_layouts import RegionGeometry, SpeakerLayout, get_speaker_layout
from trim_balance import apply_balance, balance_layers, channel_gains
Q15_SCALE = 32768.0 Q15_SCALE = 32768.0
Q15_MAX = 32767.0 / Q15_SCALE Q15_MAX = 32767.0 / Q15_SCALE
@@ -34,8 +35,9 @@ def bed_object_targets(bed_channels) -> dict[int, tuple[float, float, float]]:
def _all_object_targets(layout: SpeakerLayout, state: JocFieldState, def _all_object_targets(layout: SpeakerLayout, state: JocFieldState,
bed_targets=None) -> np.ndarray: bed_targets=None, balance=(0.0, 0.0)) -> np.ndarray:
bed_targets = bed_targets or {} bed_targets = bed_targets or {}
listener, overhead = balance
result = np.zeros((15, layout.channel_count), dtype=np.float64) result = np.zeros((15, layout.channel_count), dtype=np.float64)
for object_index in range(1, 16): for object_index in range(1, 16):
fixed = bed_targets.get(object_index) fixed = bed_targets.get(object_index)
@@ -45,6 +47,7 @@ def _all_object_targets(layout: SpeakerLayout, state: JocFieldState,
u = state.q[(object_index, "q1")] / Q15_SCALE u = state.q[(object_index, "q1")] / Q15_SCALE
v = state.q[(object_index, "q2")] / Q15_SCALE v = state.q[(object_index, "q2")] / Q15_SCALE
w = state.q[(object_index, "q3")] / Q15_SCALE w = state.q[(object_index, "q3")] / Q15_SCALE
v = apply_balance(v, w, listener, overhead)
result[object_index - 1] = render_point_gains( result[object_index - 1] = render_point_gains(
layout, u, v, w, standard_order=False) layout, u, v, w, standard_order=False)
return result return result
@@ -210,12 +213,15 @@ class PythonSpeakerRenderer:
"""Stateful float64 speaker renderer.""" """Stateful float64 speaker renderer."""
def __init__(self, layout: str | SpeakerLayout, block_size: int = 32, def __init__(self, layout: str | SpeakerLayout, block_size: int = 32,
bed_channels=None): bed_channels=None, trim_meta=None):
self.layout = get_speaker_layout(layout) if isinstance(layout, str) else layout self.layout = get_speaker_layout(layout) if isinstance(layout, str) else layout
self.block_size = int(block_size) self.block_size = int(block_size)
if self.block_size <= 0: if self.block_size <= 0:
raise ValueError("block_size must be positive") raise ValueError("block_size must be positive")
self.bed_targets = bed_object_targets(bed_channels) self.bed_targets = bed_object_targets(bed_channels)
self.trim_gains = np.asarray(channel_gains(trim_meta, self.layout),
dtype=np.float64)
self.balance = balance_layers(trim_meta, self.layout)
self.reset() self.reset()
def reset(self): def reset(self):
@@ -230,7 +236,8 @@ class PythonSpeakerRenderer:
def _apply_payload(self, payload): def _apply_payload(self, payload):
update = frame_update(payload) update = frame_update(payload)
self._state.apply(update["values"]) self._state.apply(update["values"])
new_target = _all_object_targets(self.layout, self._state, self.bed_targets) new_target = _all_object_targets(self.layout, self._state, self.bed_targets,
self.balance)
blocks = ramp_block_count(update["ramp_duration_samples"], self.block_size) blocks = ramp_block_count(update["ramp_duration_samples"], self.block_size)
difference = new_target - self._current difference = new_target - self._current
significant = np.abs(difference) >= GAIN_SNAP_THRESHOLD significant = np.abs(difference) >= GAIN_SNAP_THRESHOLD
@@ -299,6 +306,7 @@ class PythonSpeakerRenderer:
for payload in events_by_block.get(total_blocks, ()): for payload in events_by_block.get(total_blocks, ()):
self._apply_payload(payload) self._apply_payload(payload)
output *= self.trim_gains[None, :]
if standard_order: if standard_order:
return output[:, list(self.layout.internal_to_standard)] return output[:, list(self.layout.internal_to_standard)]
return output return output
+181
View File
@@ -0,0 +1,181 @@
"""Programme trim and front-back balance taken from the OAMD trim element.
TS 103 420 splits a programme's trim data into nine configurations, selected
by a layout's surround and height speaker counts:
cfg = kappa_f(surrounds) + 3 * kappa_h(heights)
kappa_f(n) = 0 for n == 0, 1 for n in (2, 3), 2 otherwise
kappa_h(n) = 0 for n == 0, 1 for n == 2, 2 otherwise
Each configuration holds one unit of calibration:
* a level in dB for the centre, surround and height speaker groups, and
* a front-back balance for the listener plane and for the overhead layer.
Only the layouts listed in ``SUPPORTED_LAYOUTS`` carry these semantics.
Configurations left at default contribute neither a static level nor a balance:
the automatic attenuation is already applied per object by the layout position
compensation. Configurations declared as disabled contribute nothing either.
The balance reparameterises the front-back axis; it does not move objects off
the axis and it leaves the two walls untouched. See ``front_back``.
"""
from __future__ import annotations
# Layouts that carry trim and front-back balance semantics, by their compact
# names: 2.0, 3.1, 4.0, 5.1, 5.1.2, 5.1.4 and 7.1. Wider layouts leave the
# programme untrimmed.
SUPPORTED_LAYOUTS = ("20", "31", "40", "51", "512", "514", "71")
# Channel label -> trim group. Left, right and LFE carry no trim in any
# configuration, so they are absent from the table.
GROUPS = {
"C": "centre",
"Ls": "surround", "Rs": "surround",
"Lrs": "surround", "Rrs": "surround",
"Ltf": "height", "Rtf": "height",
"Ltr": "height", "Rtr": "height",
"Ltm": "height", "Rtm": "height",
}
FLOOR_LABELS = ("Ls", "Rs", "Lrs", "Rrs")
HEIGHT_LABELS = ("Ltf", "Rtf", "Ltr", "Rtr", "Ltm", "Rtm")
# bal3d_y is coded in steps of 1/16, so a balance is quantised to that grid.
BALANCE_STEPS = 16
# Front-back balance keeps a 5 percent wall band at each end and leaves it unmapped.
WALL = 0.05
def _height_class(count):
if count == 0:
return 0
if count == 2:
return 1
return 2
def _floor_class(count):
if count == 0:
return 0
if count in (2, 3):
return 1
return 2
def layout_classes(layout):
"""Return ``(config_index, height_count, floor_count)`` for a layout."""
channels = tuple(getattr(layout, "channels", ()))
heights = sum(1 for c in channels if c in HEIGHT_LABELS)
floor = sum(1 for c in channels if c in FLOOR_LABELS)
return _floor_class(floor) + 3 * _height_class(heights), heights, floor
def _supported(layout):
return getattr(layout, "name", None) in SUPPORTED_LAYOUTS
def _record(atmos_meta, layout):
if not atmos_meta or not _supported(layout):
return None
index, _, _ = layout_classes(layout)
configs = atmos_meta.get("configs") or ()
if index >= len(configs):
return None
return configs[index]
def group_levels_db(atmos_meta, layout):
"""Trim level in dB for the centre, surround and height speaker groups.
Only an explicit configuration contributes a static level. An automatic
configuration carries no static gain: the attenuation it describes is
already applied per object by the layout position compensation, and
multiplying it in a second time would count it twice.
"""
record = _record(atmos_meta, layout)
if not isinstance(record, dict) or record.get("default") or record.get("disabled"):
return {"centre": 0.0, "surround": 0.0, "height": 0.0}
return {
"centre": float(record.get("centre", 0.0)),
"surround": float(record.get("surround", 0.0)),
"height": float(record.get("height", 0.0)),
}
def channel_gains(atmos_meta, layout):
"""Per-channel linear gain for the layout, in channel order."""
levels = group_levels_db(atmos_meta, layout)
gains = []
for label in getattr(layout, "channels", ()):
group = GROUPS.get(label)
db = levels[group] if group else 0.0
gains.append(10.0 ** (db / 20.0))
return gains
def balance_layers(atmos_meta, layout):
"""Front-back balance for the listener plane and the overhead layer.
Both are returned as signed magnitudes in [-1, 1]; zero means no balance.
Only an explicit configuration carries a balance.
"""
record = _record(atmos_meta, layout)
if not isinstance(record, dict) or record.get("default") or record.get("disabled"):
return 0.0, 0.0
listener = float(record.get("bal3d_y_lis", 0.0))
overhead = float(record.get("bal3d_y_tb", 0.0))
return listener, overhead
def _curve(balance):
"""Segment coefficients for one balance value.
Returns ``(a0, a1, a2, a3)`` for the two halves of the axis, or ``None``
when the balance is zero.
Writing ``s = |n| / 18`` with ``n`` the balance quantised to the coding
grid, a positive balance stretches the front half by ``1 + s`` and
compresses the back half by ``1 - s``; a negative balance does the
reverse. Both halves anchor at the axis midpoint and at the two walls,
so the mapped value always stays inside the wall range.
"""
n = int(round(abs(balance) * BALANCE_STEPS))
if n == 0:
return None
s = n / 18.0
if balance > 0:
return 1.0 + s, -WALL * s, 1.0 - s, 0.5 + 0.45 * s
return 1.0 - s, WALL * s, 1.0 + s, 0.5 - 0.45 * s
def front_back(position, balance):
"""Reparameterise a front-back coordinate in [0, 1] by ``balance``.
``0`` is the front wall and ``1`` the back wall. Positions inside the
wall bands are returned unchanged, and so is every position when the
balance is zero.
"""
curve = _curve(balance)
if curve is None:
return position
if position <= WALL or position >= 1.0 - WALL:
return position
a0, a1, a2, a3 = curve
if position < 0.5:
return a0 * position + a1
return a2 * (position - 0.5) + a3
def apply_balance(position, height, listener, overhead):
"""Blend the two balance layers by height and map the coordinate.
``height`` runs from 0 at the listener plane to 1 at the top layer.
"""
if not listener and not overhead:
return position
low = front_back(position, listener)
high = front_back(position, overhead)
return (1.0 - height) * low + height * high