Ambisonics NPM

JSAmbisonics

A JS library for first-order ambisonic (FOA) and higher-order ambisonic (HOA) processing for browsers, using the Web Audio API. Live Demo

Archontis Politis (Aalto University)
archontis.politis@aalto.fi
David Poirier-Quinot (IRCAM)
david.poirier-quinot@ircam.fr

Description

JSAmbisonics is a JavaScript library that implements a set of objects for real-time spatial audio processing, using the Ambisonics framework. The objects correspond to typical ambisonic processing blocks, and internally implement Web Audio graphs for the associated operations.

The library is suitable for both FOA and HOA processing, using the following specifications:

FOA or HOA with ACN channel ordering and N3D normalization. This is the default mode, and all internal processing is done using this.
FOA or HOA with ACN channel ordering and SN3D normalization.
FOA or HOA (up to 3rd-order) using the Furse-Malham specification.
FOA B-format signals with traditional ordering W X Y Z and a factor of 1/sqrt(2) on the omnidirectional first channel.

The implemented Web Audio classes are:

monoEncoder: encodes a monophonic sound source to an ambisonic stream of a set order, with real-time control of the panning direction.
sceneRotator: rotates the sound scene of an ambisonic stream, with real-time control of yaw, pitch, and roll rotation angles.
sceneMirror: mirrors the sound scene of an ambisonic stream with respect to (front-back), (left-right), or (up-down) axes.
virtualMic: applies FOA and HOA virtual microphones to an ambionic stream, with real-time control of their orientation and pattern.
binDecoder: implements an ambisonic to binaural decoding, using user-defined HRTF-based filters. If these are not provided, two plain opposing cardioids are used instead.
orderLimiter: takes a HOA stream of order N, and outputs the channel-limited HOA stream of order N'<=N
orderWeight: applies user-specified gains to the channels of the same order, for directional smoothing or psychoacoustic (max energy-vector) decoding
converters.wxyz2acn: converts a traditional FOA stream to FOA ACN/N3D stream
converters.acn2wxyz: converts the first-order channels of a HOA stream to traditional FOA stream
converters.n3d2sn3d: converts an ACN/N3D stream to an ACN/SN3D stream
converters.sn3d2n3d: converts an ACN/SN3D stream to an ACN/N3D stream
converters.fuma2acn: converts a FuMa stream to a ACN/N3D stream
intensityAnalyser: implements an acoustic intensity analysis for visualization of directional information captured in the ambisonic stream
powermapAnalyser: analysis of the directional power distribution in the sound field for visualization of directional information captured in the ambisonic stream
rmsAnalyser: returns the RMS values of the ambisonic channels, useful for metering and visualization.
convolver: performs multichannel convolution, typically used to convolve a mono anechoic file with an ambisonic room impulse response.

The library is a work-in-progress, but fully functional. At the moment, demos seem to work fine in Mozilla Firefox and Google Chrome. Safari seems to be working too with some issues in multichannel file loading. No other browsers have been checked yet.

If you would like to reference the library in an article please use the following publication:

JSAmbisonics: A Web Audio library for interactive spatial sound processing on the web
A. Politis, D. Poirier-Quinot
Interactive Audio Systems Symposium, York, UK, 2016

in which you can also find a detailed description of the internals of the library.

Computation of spherical harmonics and rotations relies on the JavaScript spherical harmonic library contributed by the author here. The HOA code is based on the larger Matlab HOA and Spherical Harmonic Transform libraries contributed by the author in Github. The rotation algorithm is the fast recursive one by Ivanic and Ruedenberg.

Real-time demos (Chrome and Firefox)

See the live Rawgit demo (serving the content of the ./examples folder).

HOA recordings are made by the author in the Communication Acoustics laboratory of Aalto University, using the Eigenmike microphone.

Projects using JSAmbisonics

We are aware of the following projects using the library:

Web Audio spatial scene rendering with interactive walkthrough capabilities (Thomas Deppisch, IEM Graz)
Veri: a 360 video player
Patches: the browser-based VR world-making platform of Vizor
Mobile spatial audio experiments by Infinite Dimensions Audio

Installation and usage

To add the library to you node project, type in (terminal at project root):

npm install ambisonics

To use the ambisonic objects, include the JSAmbisonics library in the body of your html code as:

<script type="text/javascript" src="ambisonics.umd.js"></script>

or use it directly as a node module:

var ambisonics = require('ambisonics');

ambisonic encoder is initialized as

var encoder = new ambisonics.monoEncoder(audioContext, order)

where audioContext is the current Web Audio context, and order the desired ambisonic order. The input stream comes from an audio node to be spatialized. The azimuth and elevation of the encoded source can be updated at runtime by

encoder.azim = azim_value_in_degrees;
encoder.elev = elev_value_in_degrees;
encoder.updateGains();

ambisonic mirror is initialized as

var mirror = new ambisonics.sceneMirror(audioContext, order)

The reflection planes (1: front-back), (2: left-right), (3: up-down), or (0: no reflection) can be updated at runtime by

mirror.mirror(planeNo);

where planeNo is any of the above integers.

ambisonic rotator is initialized as

var rotator = new ambisonics.sceneRotator(audioContext, order)

The yaw (Z-axis rotation), pitch (Y-axis) and roll (X-axis) rotation angles can be updated at runtime by

rotator.yaw = yaw_value_in_degrees;
rotator.pitch = pitch_value_in_degrees;
rotator.roll = roll_value_in_degrees;
rotator.updateRotMtx();

ambisonic binaural decoder is initialized as

var binDecoder = new ambisonics.binDecoder(audioContext, order)

If no decoding filters are passed to the decoder, an initial decoding based on two opposing cardioids is defined by default. In case binaural decoding filters are available, they should be loaded in a multichannel audioBuffer and passed to the decoder through

binDecoder.updateFilters(audioBuffer);

The number of channels of the buffer should be equal or greater to (order+1)^2, which amounts to the number of ambisonic channels for the specified order. The filters can be reset to their default cardioids by

binDecoder.resetFilters();

Some FOA and HOA HRTF-based decoding filters are included in the examples.

ambisonic virtual microphone is initialized as

var vmic = new ambisonics.virtualMic(audioContext, order)

The virtual microphone is initialized to a hypercardioid of the appropriate order, pointing to the front. The orientation can be updated at runtime by

vmic.azim = azim_value_in_degrees;
vmic.elev = elev_value_in_degrees;
vmic.updateOrientation();

The pattern of the microphone can be also updated by

vmic.vmicPattern = string;
vmic.updatePattern();

where string can be one of the following:

"cardioid"
"supercardioid"
"hypercardioid"
"max_rE"

Higher-order cardioids correspond to the normal cardioid raised to the power of order. Higher-order supercardioids correspond to the pattern of that order that maximizes the front-to-back energy ratio. Higher-order hypercardioids correspond to the pattern of that order that maximizes the directivity factor. the max-rE pattern, found in ambisonic decoding literature, corresponds to the pattern of that order that maximizes the (Gerzon) energy vector for diffuse sound.

All objects have an input node object.in and an output node object.out which are used for the connections. The number of channels expected from each input depends on the object (for example a FOA encoder expects a monophonic signal, and outputs 4 channels, a FOA binaural decoder expects 4-channel input and outputs 2 channels, etc.). Example connections:

soundBufferPlayer.connect(encoder.in);
encoder.out.connect(rotator.in);
rotator.out.connect(binDecoder.in);
binDecoder.connect(audioContext.destination);

which implements a graph such as:

soundBufferPlayer ------------->encoder------------>rotator---------->binDecoder-------------->out
                  mono stream            HOA stream         HOA stream            stereo stream

and

soundBufferPlayer.connect(vmic.in);
vmic.out.connect(audioContext.destination);

with a graph such as:

soundBufferPlayer ------------->vmic------------>out
                  HOA stream         mono stream

See the scripts in the ./examples folder for more insights on how to use the different objects of the library.

Loading of multichannel files for HOA

The HOA processing of order=N requires audio streams of (N+1)^2 channels. Loading HOA recordings or HOA binaural filters from sound files of that many channels seems to be problematic for the browsers. Both Firefox and Chrome seem to be able to handle WAVE and OGG files of up to 8ch. For that reason a helper class is provided that loads individual 8ch files that have been split from the full HOA multichannel file. Usage:

var HOA3soundBuffer;
var order = 3;
var url = "https://address/HOA3_rec1.wav";
var callbackOnLoad = function(mergedBuffer) {
    HOA3soundBuffer = mergedBuffer;
}
var HOA3loader = new ambisonics.HOAloader(audioContext, order, url, callbackOnLoad);
HOA3loader.load();

The class will try to find files with the provided file name HOA3_rec1.wav but of the form:

  HOA3_rec1_01-08ch.wav
  HOA3_rec1_09-16ch.wav

The above example for 3rd-order will have exactly two files of 8ch (16 HOA channels). For a 2nd-order example (9 HOA channels) the loader will check for

  HOA2_rec1_01-08ch.wav
  HOA2_rec1_09-09ch.wav

and so on.

Integration with SOFA HRTFs

Generation of the binaural decoding filters require Head-related Transfer Functions (HRTFs), and they depend on the order, the measurement grid of the HRTFs, and the choice of ambisonic decoding approach. This is often not a straightforward task and not for the end-user of the library. If users are not satisfied with the example filters provided in the repo, and they have access to various HRTF sets, or to their own personalized HRTFs, they should have an automated way to generate these filters. A module that loads HRTFs and attempts to do that is included in the library. The HRTFs should be in the standardized SOFA format, which seems to be the most established one at the moment. Usage:

var decodingFilterBuffer;
var order = 3;
var url = "https://address/HRTFset_subjectXX.sofa.json";
var callbackOnLoad = function(decodedBuffer) {
    decodingFilterBuffer = decodedBuffer;
}

var binDecoder = new ambisonics.binDecoder(audioContext, order);
var hrirLoader = new ambisonics.HRIRloader_xxxxx(audioContext, order, callbackOnLoad);
hrirLoader.load(url);
binDecoder.updateFilters(decodingFilterBuffer);

The SOFA file should be first converted to a JSON file. The appropriate HRIRloader_xxxxx should be used, depending on the origin of the SOFA files.

HRIRloader_ircam

This loader is meant to be used with JSON-converted SOFA files from IRCAM. Two examples are included in the repo. This loader relies on the IRCAM module serveSofaHrir and it is included for future loading of HRTFs served publicly from IRCAM. Scripts to convert your own SOFA files to this JSON convention are not included.

HRIRloader_local

This loader is meant to be used with the user's own SOFA files. You need to have Python installed to convert the SOFA files to JSON, and the h5py Python library installed. Using PIP do:

pip install h5py

Then in the ./utils folder do (thanks to Antti Vanne for the script)

python sofa2json.py [HRTFsetFilePath].sofa

That will generate the same file but as HRTFsetFilePath.sofa.json, which you can then load normally, using the loader as above.

2D HOA processing

In many cases a scene may be limited to sounds coming only from the horizontal plane, either due to scene properties (e.g. a cocktail party of multiple speakers at ear level), due to rendering restrictions (e.g. HRTF filters available only on the horizontal plane), or due to reduced processing requirements. In such cases not all HOA components are active, since sounds are constrained only to the horizontal plane, without elevation information. For HOA processing of order=N, then only (2N+1) HOA channels are active. To take advantage of the reduced processing and bandwidth needs of 2D HOA processing, it is better to design the encoder, rotation, decoding, and rest of the modules from scratch instead of just using the full 3D ones. Such objects have been contributed by Thomas Deppisch, developed as part of his bachelor thesis at Institute of Electronic Music and Acoustics (IEM), KUG/TU Graz, Austria. Many thanks! Some 2D HOA examples are included in the demos.

The following objects have a 2D processing counterpart, their usage is exactly the same as the full 3D ones:

monoEncoder2D
sceneRotator2D
sceneMirror2D
binauralDecoder2D
orderLimiter2D
intensityAnalyser2D
hrirLoader2D_local

Legacy

In the ./legacy folder of the repository there is a copy of the FOA part of the initial release of the library, when the library was still split in FOA and HOA components. The only reason that this is preserved is that if somebody is interested in FOA only processing, the WebAudio_FOA.js file has all the components needed without the additional complexity of HOA processing and with no external dependencies.