EBU ADM Renderer (EAR)

The EBU ADM Renderer (EAR) is a complete interpretation of the Audio Definition Model (ADM) format, specified in Recommendation ITU-R BS.2076-1. ADM is the recommended format for all stages and use cases within the scope of programme productions of Next Generation Audio (NGA). This repository contains a Python reference implementation of the EBU ADM Renderer.

This Renderer implementation is capable of rendering audio signals to all reproduction systems mentioned in "Advanced sound system for programme production (ITU-R BS.2051-1)".

Further descriptions of the EAR algorithms and functionalities can be found in EBU Tech 3388.

Note: Version 2.0 of the renderer represents also the reference implementation of ITU-R BS.2127 (ITU ADM Renderer)

Test files

A initial set of ADM files to test the EAR can be found under

• https://ebu.io/qc/testmaterial and
• http://cvssp.org/data/s3a/public/radiodrama_register.php

Installation

To install the latest release from PyPi:

$ pip install ear

Python versions

EAR supports Python >=3.6 and runs on all major platforms (Linux, Mac OSX, Windows).

Installation of extra packages

If you want to run the unit tests you can install all extra requirements with pip:

$ pip install ear[test]

Getting started

The EAR reference implementation comes with two command line tools:

• ear-render Is the main tool to render BW64/ADM audio files
• ear-utils Collection of useful ADM utilities

Command line renderer

usage: ear-render [-h] [-d] -s target_system [-l layout_file]
                  [--output-gain-db gain_db] [--fail-on-overload]
                  [--enable-block-duration-fix] [--programme id]
                  [--comp-object id]
                  [--apply-conversion {to_cartesian,to_polar}] [--strict]
                  input_file output_file

EBU ADM renderer

positional arguments:
  input_file
  output_file

optional arguments:
  -h, --help            show this help message and exit
  -d, --debug           print debug information when an error occurs
  -s target_system, --system target_system
                        Target output system, accoring to ITU-R BS.2051.
                        Available systems are: 0+2+0, 0+5+0, 2+5+0, 4+5+0,
                        4+5+1, 3+7+0, 4+9+0, 9+10+3, 0+7+0, 4+7+0
  -l layout_file, --layout layout_file
                        Layout config file
  --output-gain-db gain_db
                        output gain in dB (default: 0)
  --fail-on-overload, -c
                        fail if an overload condition is detected in the
                        output
  --enable-block-duration-fix
                        automatically try to fix faulty block format durations
  --programme id        select an audioProgramme to render by ID
  --comp-object id      select an audioObject by ID from a complementary group
  --apply-conversion {to_cartesian,to_polar}
                        Apply conversion to Objects audioBlockFormats before
                        rendering
  --strict              treat unknown ADM attributes as errors

To render an ADM file, the following three parameters must be given:

• -s followed by the target output format to render to
• the name of the input file
• the name of the output file

For example ear-render -s 0+5+0 input.wav output_surround.wav will render the BW64/ADM file input.wav to a 0+5+0 target speaker layout and store the result in output_surround.wav.

The optional --layout parameter allows to specify the real loudspeaker positions and screen dimensions of a reproduction setup. Refer to the layout file documentation for more information about its format.

--fail-on-overload makes the rendering process fail in case an overload in the output channels to ensure any signal clipping doesn't go unnoticed. Use --output-gain-db to adjust the output gain.

--enable-block-duration-fix automatically fixes durations of audioBlockFormats in case they are not continuous. Please note that the proper way to handle this situation is to fix the input file.

--strict enables strict ADM parsing mode. Some of the currently available ADM/BW64 files may not strictly adhere to the BS.2076 specification, for example by including xml attributes that are not part of the standard. The default behaviour is to output a warning and continue processing. When strict mode is enabled, warnings are turned into errors and processing is stopped.

Please note that, depending on the size of the file, it may take some time to render the file. At the time of writing, the parsing of the ADM XML data is relatively slow when the ADM is large (>= a few megabytes).

Command line ADM utilities

The ear-utils command provides various subcommands which can be seen on the help message when called with ear-utils --help:

Each subcommand may have its own command line options, which can be displayed using ear-utils SUBCOMMAND --help, where SUBCOMMAND is one of the supported subcommands.

usage: ear-utils [-h]
                 {make_test_bwf,replace_axml,dump_axml,dump_chna,ambix_to_bwf}
                 ...

EBU ADM renderer utilities

optional arguments:
  -h, --help            show this help message and exit

available subcommands:
  {make_test_bwf,replace_axml,dump_axml,dump_chna,ambix_to_bwf}
    make_test_bwf       make a bwf file from a wav file and some metadata
    replace_axml        replace the axml chunk in an existing ADM BWF file
    dump_axml           dump the axml chunk of an ADM BWF file to stdout
    dump_chna           dump the chna chunk of an ADM BWF file to stdout
    ambix_to_bwf        make a BWF file from an ambix format HOA file

HOA ADM Creation

usage: ear-utils ambix_to_bwf [-h] [--norm NORM] [--nfcDist NFCDIST]
                              [--screenRef] [--chna-only]
                              input output

positional arguments:
  input              input file
  output             output BWF file

optional arguments:
  -h, --help         show this help message and exit
  --norm NORM        normalization mode
  --nfcDist NFCDIST  Near-Field Compensation Distance (float)
  --screenRef        Screen Reference
  --chna-only        use only CHNA with common definitions

To convert an ambiX file in an ADM one, the following two parameters must be given:

• the name of the input file
• the name of the output file

The optional parameters are :

• The normalization of the signals (N3D, FuMa or SN3D, which is the default value)
• The NFC Distance, i.e., the distance at which the HOA mix was created. A float value between 0 and 20 meters must be given. The default value 0 means no NFC processing.
• The screenRef flag, which tells if the audio content is screen related or not. The default value is False, which means no screen scaling.

For example, ear-utils ambix_to_bwf --nfcDist 2.53 input.wav output.wav will create an ADM file called output.wav containing the audio samples of the input.wav file and the ADM metadata corresponding to an ambiX file with SN3D normalization, an 2.53 meters nfcDist, and no screen scaling.

Please note that the software implicitly assumes that all the HOA channels are in ACN ordering and that no channel is missing. For example, it will assumes the signal is a 4th order HOA signal if it finds 25 channels ((4+1)²=25).

Changelog

2.1.0 - 2022-01-26

Fixed

• Depth and height parameters were switched in metadata conversion. See #26.
• Bug in channel lock priority order, which controls the loudspeaker selection when the object position is the same distance from multiple loudspeakers. See #28.
• Screen scaling now fails explicitly in cases where it was not well-defined before, generally with extreme positions and sizes. See #22.
• Errors with gaps at the start of metadata. See #13.
• Rounding of times in XML writer. See #12.
• audioStreamFormat referencing error messages. See 34b738a and 04533fc.
• Improved extraData handling in BW64 reader; see #48

Changed

• DirectSpeakers panner uses allocentric panning for Cartesian positions. See 222374a.
• Removed python 2.7 support.
• fix_block_format_durations parameter is deprecated, and the ADM XML parser no longer issues warnings for timing issues -- use ear.fileio.adm.timing_fixes for this functionality instead. See #8.
• --enable-block-duration-fix performs more extensive fixes; this now fixes the following issues:
  • audioBlockFormats where the rtime plus the duration of one audioBlockFormat does not match the rtime of the next.
  • interpolationTime parameter larger than duration.
  • audioBlockFormat rtime plus duration extending past the end of the containing audioObject.
• Issue a warning for DirectSpeakers blocks with a speakerLabel containing LFE which is not detected as an LFE channel. See #9.
• Improved warning and error output: tidier formatting, and repeated warnings are suppressed by default. See #37.

Added

• loudnessMetadata data structures, parsing and generation. See #25.
• ear-utils regenerate command to re-generate AXML and CHNA chunks. See #8.
• The absoluteDistance parameter is now extracted from AXML and added to the ExtraData structure; see #45.
• Lots of documentation, see https://ear.readthedocs.io/

2.0.0 - 2019-05-22

Changes for ITU ADM renderer reference code.

• Changes to rendering of Objects specified using Cartesian parameters.
• Added conversion between Cartesian and polar Objects parameters.
• Changed to BSD-Clear license.

1.2.0 - 2019-05-22

Fixed

• Slightly non-normalised gains in extent panner.

Changed

• Improved downmix/upmix behaviour for DirectSpeakers content specified using the common definitions.
• Change default azimuth of M+-SC to 15.
• Support wider azimuths for M+-SC; they may now be between 5 and 25 degrees, or 35 and 60 degrees.
• Removed LFE behaviour for Objects.
• Validate Cartesian and frequency information for Objects

1.1.2 - 2019-05-22

Fixed

• Bug whereby NaNs could be produced in polar extent panner.

1.1.1 - 2019-04-12

Fixed

• aspectRatio is an attribute not an element.
• Accept the urn:ebu:metadata-schema:ebuCore_2016 namespace.
• Added default for Z coordinates.
• Various deprecation warnings.

Added

• Specified loudspeaker positions are now checked against the allowed ranges.
• Validation of screenEdgeLock in Objects.

1.1.0 - 2018-11-26

Removed

• adm_parent references in ADM objects.

Added

• Matrix type support.
• Simple validation of ADM data model, and more extensive validation of ADM structures.
• Support for rendering objects with silent audioTrackUID references.
• Support for audioPackFormat HOA parameters.
• Selection of audioProgramme and complementary audioObjects from ear-render.
• adm_path to RenderingItems, with pointers to the corresponding ADM objects. This is not used by the renderer but is a useful feature for other applications built on top of the underlying ADM library.
• Support for more flexible referencing of nested audioPackFormats; each audioTrackUID or chna row can reference any appropriate audioPackFormat on the path from the root audioPackFormat (which contains the full set of audioChannelFormats used) to the audioPackFormat that contains the audioChannelFormat of the track.
• Better support for using multi-channel audioPackFormats -- in an audioObject or chna-only file using multiple audioPackFormats may be ambiguous if they share some audioChannelFormats. This should now be handled correctly in all cases.

Changed

• The reference direction between audioTrackFormat and audioStreamFormat was reversed in the data model (audioTrackFormats now point at a single audioStreamFormat), and axml references in either direction now establish this relationship. Note that this does not follow the exact wording in BS.2076-1, but this helps compatibility with other systems and should match future revisions of BS.2076. If either of these references are omitted a warning will be issued when a file is rendered. When generating BW64+ADM files (with ear-utils) both reference directions are now included.
• Complete re-implementation of RenderingItem selection to support other features in this release. This functionality was moved from ear.fileio.utils to ear.core.select_items.
• RenderingItems now use the TrackSpec structure rather than an index to point to their source audio, to allow for silent and Matrix tracks.

1.0.1 - 2018-11-23

Changed

• An error will be raised if any audioTrackUID elements or CHNA rows have ID ATU_00000000, as references to these elements could be confused with silent audioTrackUID references.
• Miscellaneous improvements to testing infrastructure.

Fixed

• When generating BW64+ADM files (with ear-utils), do not generate audioTrackUIDs with ID ATU_00000000, which (as above) could be confused with silent audioTrackUID references.
• Updated use of attrs to fix deprecation warnings.
• Wrong imports and CHNA chunk generation in replace_axml command.
• Pytest warnings fixed by upgrading pytest-datafiles to 2.0.
• Error when testing block_aligner on python 3.7 with coverage enabled.
• Error in PeakMonitor when rendering very short files.
• dump_chna in binary mode on python 3.
• Padding character in axml chunk.

1.0.0 - 2018-03-29

Initial release.