A pure Rust implementation for reading HRTF filters from SOFA files
(Spatially Oriented Format for Acoustics).
The render module implements uniformly partitioned convolution algorithm
for rendering HRTF filters.
Based on the libmysofa C library by Christian Hoene / Symonics GmbH.
use sofar::reader::{Filter, OpenOptions};
use sofar::render::{Renderer, RendererPlan};
// Open sofa file, resample HRTF data if needed to 44_100
let sofa = OpenOptions::new()
.sample_rate(44100.0)
.open("my/sofa/file.sofa")
.unwrap();
let filt_len = sofa.filter_len();
let mut filter = Filter::new(filt_len);
// Get filter at position
sofa.filter(0.0, 1.0, 0.0, &mut filter);
let plan = RendererPlan::builder(filt_len)
.with_sample_rate(sofa.sample_rate())
.with_partition_len(64)
.build()
.unwrap();
let mut render = Renderer::new(&plan);
render.set_filter(&filter).unwrap();
let input = vec![0.0; 256];
let mut left = vec![0.0; 256];
let mut right = vec![0.0; 256];
// read_input()
render.process_block(&input, &mut left, &mut right).unwrap();Filter delays, which some SOFA files use for interaural time differences, are
ignored unless enabled with
Renderer::builder(&plan).with_max_delay(sofa.max_delay()). For several
sources, create one renderer per source from the same plan and mix them with
process_block_add.
You can run cpal renderer example like this:
cargo run --example renderer -- <FILENAME-MONO.wav> libmysofa-sys/libmysofa/share/default.sofaset_filter prepares filters on the calling thread. To keep that work out of
the audio callback, split a renderer with into_realtime: a worker thread
publishes filters, and the audio thread adopts them at partition boundaries
without allocating or freeing memory. Continuing the first example:
use sofar::render::{FilterTransition, Renderer};
let (mut publisher, mut render) = Renderer::builder(&plan)
.with_filter_transition(FilterTransition::crossfade(32))
.with_max_delay(sofa.max_delay())
.build()
.unwrap()
.into_realtime();
// Worker thread, whenever the source moves:
sofa.filter(1.0, 0.0, 0.0, &mut filter);
publisher.publish_filter(&filter).unwrap();
// Audio callback:
render.process_block(&input, &mut left, &mut right).unwrap();- Latest wins: publishing replaces a filter the renderer has not adopted yet; a running crossfade finishes before the next one starts.
- Validation: invalid filters are rejected on the worker and never reach
the renderer.
publishsends already prepared filters, such as cached ones. - Memory: displaced filters return to the publisher, which frees them when
publishing or on
reclaim(). - Shutdown: once the renderer is dropped, publishing returns
Error::RendererDisconnected. Drop the renderer outside the audio callback; the publisher can be dropped at any time.
Custom transports can use Renderer::set_prepared_filter and
take_retired_filter, freeing the filters they return off the audio thread.
- Sample rate and partition length move to
RendererPlan::builder(filter_len); create renderers from the plan withRenderer::neworRenderer::builder. with_left_delay/with_right_delaybecomewith_max_delay(seconds), e.g.with_max_delay(sofa.max_delay()). Filters with longer delays are rejected.process_blocktakes slices;Erroris non-exhaustive with named fields.set_filterreturns an error instead of panicking on mismatched channels. Its first few calls allocate;into_realtimekeeps that off the audio thread.
This project is a Rust port of libmysofa, a C library for reading SOFA files.
- libmysofa Copyright © 2016-2017 Symonics GmbH, Christian Hoene (BSD-3-Clause)
- KD-tree Copyright © 2007-2011 John Tsiombikas (BSD-3-Clause)
See the NOTICE file for full attribution details.
This project is licensed under either of
- Apache License, Version 2.0, (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)
at your option.
