Render SOFA HRIRs directly and place bed channels by their labels
ci / ubuntu-latest (push) Failing after 11s
ci / macos-latest (push) Canceled after 0s
ci / windows-latest (push) Canceled after 0s
ci / release (push) Canceled after 0s

Two changes that belong to the same reproduction model: a bed channel has
one position, defined by its label, and a SOFA set keeps the impulse
responses it was measured with.

CBA bed channels: the position comes from the channel's label in ETSI
TS 103 420 Table B.10, not from the layout the decoder is asked to render
to. The per-layout anchor table displaced bed channels onto the nearest
target speaker, which silently moved content whenever the target layout
changed.

- Drop the anchor table and the geometry lookup; bed_channel_q15 is the
  only source of bed positions, so stream and task no longer need a
  layout to ask for one
- Keep the 17-bit non-standard bed mask in channel order, as Table 13 /
  TS 103 190-2 Table 65 transmit it
- Exclude LFE channels from the object list while keeping them in the
  coverage mask
- Cover the Table B.10 coordinates, the q15 encoding, the 10-bit and
  17-bit masks and the LFE split in tests

SOFA rendering: compiling a SOFA set into a fifth-order directional field
threw away the measured impulse response - the SH projection, the
64-QMF/77-hybrid round trip and the synthetic room replaced it. The full
IRs are now kept and rendered as such, with the reproduction defined by a
configuration that names a personalized model rather than by the renderer
itself.

- Add SphericalMesh: incremental convex hull over the measurement
  directions, barycentric weights inside the containing triangle and
  spherical Voronoi areas for the centre spectrum
- Add FullHrirField: rfft of the selected shell, 200-1500 Hz phase fit for
  the delay, delay-aligned spectra, log-magnitude/phase or MCA
  interpolation, rigid-sphere radial ratio, recording origin and the
  common FFT/2 latency
- Add Uniform/MultiSourcePartitionedConvolver: block 128, numpy FFT
  conventions, zero-fade filter rebuild from the blocks in flight
- Move the Rosella reference arithmetic (profile geometry, direction
  basis, direct gains and room send, table-A room) out of the renderer
  into rosella_reference so both renderers share it
- Add ReferenceEnvironment and SofaReferenceRuntime: JSON configuration
  with model hash and distance-profile mapping, level calibration over
  equal-area directions, the reference room, 1536-sample batching and a
  single latency removal
- Keep .jochrtf an explicit input and the SOFA compile step behind
  --hrtf-cache-policy/--hrtf-cache-dir; add --reference-config and
  --sofa-interpolation, and make --hrtf-radius-m optional for a single
  shell
- Cover the mesh weights and Voronoi areas, the sphere series and the
  partitioned convolver with reference values in tests
- Document both rules in the READMEs and the binaural notes

Rosella renders stay byte-identical (d5877e89...) and so do renders of
the legacy compiled path (6f0a75d2...). Direct SOFA renders match the
Python reference to 1e-4 in level; the residual comes from the
barycentric solve (analytic inverse here, LAPACK there) on near-degenerate
triangles.
This commit is contained in:
TheM14 committed 2026-11-01 02:54:55 +08:00
1 parent 34a722f56a
commit fd842db26f
27 files changed
+4746 -1065

No files matched your search

+107
View File
@@ -0,0 +1,107 @@
#pragma once
#include <complex>
#include <cstddef>
#include <memory>
#include <vector>
#include "foundation/status.h"
#include "hrtf/rosella_model.h"
// One of the room's transforms is a radix-2 FFT plan, so the type is named here and
// kept out of the interface.
namespace joc::dsp {
class FftPlan;
} // namespace joc::dsp
// The reference arithmetic of the Rosella ".personalized_headphone" renderer
// (rosella_model.py, rosella_direct.py and rosella_room.py): the direction basis and
// the padded lane grid, the per-source direct gains and room send, and the recursive
// table-A room model. rosella_renderer.cpp drives these through RosellaRuntime; a
// renderer built on another field format can call them directly and get the same
// samples, so the operation order and the literals in the definitions are part of
// the interface.
namespace joc::hrtf {
// rosella_direct._profile_geometry: a virtual position resolved on one distance
// profile's grid, as the direction of the direct path plus the radius factors the
// gain weights are scaled by.
struct ProfileGeometry {
double direction[3] = {1.0, 0.0, 0.0};
double radius = 0.0;
double clamped = 0.0;
double alpha = 0.0;
};
// Quantises `position` onto the profile's metadata grid, maps it through the
// profile's axis scales and clamps it to the profile bounds; `out` takes the result.
Status profile_geometry(const RosellaModel& model, const double position[3], int profile_index,
ProfileGeometry* out);
// rosella_model.direction_basis: the observed 36-term basis of one unit vector.
void direction_basis(double x, double y, double z, double out[36]);
// rosella_direct._logical_field: the padded lane grid as [77][36][2] float64.
std::vector<double> rosella_logical_field(const std::vector<float>& padded);
// One source's direct-path gains and room send, plus the radii and the interaural
// delay the caller reports. `delayed_ear` is -1 when neither ear is delayed,
// otherwise it is the ear (0 = left, 1 = right) whose phase carries
// `delay_samples`.
struct DirectResult {
std::complex<double> gains[2][77] = {};
double room_send = 0.0;
double physical_radius_m = 0.0;
double normalized_radius = 0.0;
double clamped_radius = 0.0;
double delay_samples = 0.0;
int delayed_ear = -1;
};
// rosella_direct.special_lfe_direct: the response a special (LFE) source is
// rendered with, built once.
const DirectResult& special_lfe_direct();
// rosella_direct.direct_and_room_send: the direct gains and the room send of the
// position `position` on profile `profile_index`, read through the logical field
// grids of `field_left`/`field_right`.
Status direct_and_room_send(const RosellaModel& model, const std::vector<double>& field_left,
const std::vector<double>& field_right, const double position[3],
int profile_index, DirectResult* out);
// rosella_room.RosellaRoomFir: the recursive table-A room model, its room FIR and the
// overlap-add convolution that applies the FIR in the hybrid band domain. State is
// per instance and not thread-safe.
class RosellaRoomFir {
public:
RosellaRoomFir();
~RosellaRoomFir();
RosellaRoomFir(const RosellaRoomFir&) = delete;
RosellaRoomFir& operator=(const RosellaRoomFir&) = delete;
// Builds the room impulse response from `model` over `impulse_slots` slots. The
// room can only be processed once this succeeds; the alternate table-A mode is
// rejected here, as the reference rejects it.
Status configure(const RosellaModel& model, int impulse_slots);
// `values` is [slots][2][77], `output` is [slots][2][77] and stays zero outside
// the room bands.
Status process_chunk(const std::vector<std::complex<double>>& values, std::size_t slots,
std::vector<std::complex<double>>* output);
void reset();
bool valid() const;
private:
int length_ = 0;
bool valid_ = false;
std::vector<std::complex<double>> kernel_;
std::vector<std::complex<double>> tail_;
std::vector<std::complex<double>> kernel_fft_;
std::size_t fft_cache_size_ = 0u;
std::vector<std::complex<double>> line_; // scratch, exactly fft_size entries
std::vector<std::complex<double>> input_fft_; // scratch, [fft_size][64], fully written
std::vector<std::complex<double>> block_; // scratch, [needed][2][64], fully written
std::unique_ptr<dsp::FftPlan> forward_plan_;
std::unique_ptr<dsp::FftPlan> inverse_plan_;
};
} // namespace joc::hrtf