annotate CHANGELOG @ 309:12eadc54e874

Start rejigging README

Changes in Sonic Annotator 1.5 since the previous release 1.4:

Front-end changes:

 - Change mp3 file loading so as to compensate for encoder/decoder
   delay and padding (using "gapless playback" logic). While this is a
   real improvement, unfortunately it does mean that the initial
   padding in mp3 file load has changed from previous versions, which
   may change alignment of features based on mp3 files relative to
   those generated by previous versions. It's not a great idea to use
   mp3 as a file format of record, because of differences like this
   between decoders

 - Add --transform-minversion flag to test for available version of
   a plugin transform
 
 - Add quiet mode (-q)

Bug fixes:

 - Fix possible crash in multiplexed file handling (--multiplex, -m)

 - Fix nominal rounding error (out by 1ns) in output under some
   conditions


Changes in Sonic Annotator 1.4 since the previous release 1.3:

Front-end changes:

 - Better error reporting, especially for invalid transform files
   and transform-not-found

 - Avoid crashing out when a single plugin (that is not being used)
   can't be loaded because of e.g. an undefined symbol

Bug fixes:

 - Fix (with test) horrible crash with --multiplex option

 - Fix erroneous quantization to 16 bits for coded file types of
   greater bit depth

 - Fix multiple outputs when requesting both summary and non-summary
   for the same output


Changes in Sonic Annotator 1.3 since the previous release 1.2:

Back-end (feature writer) changes:

 - Add the --csv-digits, --lab-digits, and --jams-digits options to
   control the number of significant figures printed for feature values

 - Update the JAMS writer to JAMS 0.2.0 format

 - Show a more useful error message when user provides a filename
   instead of a transform id to the -d option (happened to me when
   cut-and-pasting command lines and I was quite baffled at first)


Changes in Sonic Annotator 1.2 since the previous release 1.1:

Front-end changes:

 - Add the --segments-from option, providing the ability to read
   segment boundaries from a file

Back-end (feature writer) changes:

 - Rename the JSON feature writer to JAMS. There may be other JSON
   formats supported in future

Bug fixes:

 - Fix invalid JSON written by JAMS feature writer for dense
   features.
 - Fix invalid UTF-8 output from RDF feature writer when processing
   MP3 files having ID3 tags in non-ASCII, non-UTF8 encodings


Changes in Sonic Annotator 1.1 since the previous release 1.0:

Front-end changes:

 - Add support for the start time and duration properties of a
   transform, applying a plugin to only a range of the input audio
 - Reduce the internal processing blocksize from 16384 to 1024 samples
   to reduce extent of time rounding at end of file or range
 - Add --multiplex option to compose multiple audio files into a
   single multi-channel stream with one input file per channel
 - Add --normalise to request each audio file be normalised to 1.0 max
 - Add support for the plugin_version property of a transform, causing
   Sonic Annotator to refuse to run with the wrong version of a plugin
 - Add --minversion option to permit scripts to check that the
   version of Sonic Annotator is as they expect
 - Add new housekeeping options to list the available feature writers
   and supported audio file formats
 - Pull out the feature-writer-specific help text into separate help
   options (-h <writertype>) as the help was getting too long

Back-end (feature writer) changes:

 - Add --csv-omit-filename, --csv-end-times, and --csv-fill-ends
   options to the CSV feature writer to adjust various aspects of its
   output
 - Add "json" feature writer, exporting to JAMS (JSON Annotated Music
   Specification) format. This writer is provisional and is expected
   to change in future releases to comply more effectively with the
   specification
 - Add "midi" feature writer, exporting to MIDI files
 - Add "lab" feature writer, exporting to tab-separated label files.
   (This is equivalent to using the CSV writer with a tab separator and
   the new --csv-omit-filename and --csv-end-times options, but it's
   simpler to use if .lab is what you want)

Bug fixes:

 - Fix the former habit of forging ahead even if not all transform
   files could be found or parsed (this may have been intentional
   behaviour but it is confusing more than it is useful)
 - Fix failure to support --summary-only flag when reading transforms
   with summaries from a transform file


Changes in Sonic Annotator 1.0 since the previous release 0.7:

Bug fixes:

 - Fix incorrect samplerate in reading m4a files on OS/X
 - Fix incorrect handling of FixedSampleRate outputs (Vamp SDK fix)
 - Add tests that use the Vamp test plugin


Changes in Sonic Annotator 0.7 since the previous release 0.6:

Build changes:

 - Support 64-bit builds on OS/X (using CoreAudio instead of
   obsolete QuickTime audio file reader)

 - Simplify RDF reading and fix some bugs. Now requires Dataquay
   (http://breakfastquay.com/dataquay/) rather than using Redland
   directly. Release builds use Sord/Serd rather than Redland


Changes in Sonic Annotator 0.6 since the previous release 0.5:

Build changes:

 - Switch to modular SV-libraries build using svcore library
 - We now require Vamp plugin SDK v2.3
 - Add autoconf configure script

Front-end changes:

 - Pick up default sample rate and channel count from the first
 audio file, where not specified in the transform, instead of using
 hard coded defaults
 - Make it possible to specify the window shape in transform
 - Fix the --csv-one-file option which did not work in 0.5
 - Fix --force option when using playlists
 - Add -v option to print version number and exit


Changes in Sonic Annotator 0.5 since the previous release 0.4:

Build changes:

 - Remove unused audioio library and avoid its dependencies

Front-end changes:

 - Avoid leaking file descriptors on exceptions
 - Fix embarrassing bug that caused failure to mix down to mono
 properly for single-channel plugins with multi-channel input files
 - Fail sooner if the output file is not writable (i.e. don't wait
 until the first data is available for writing)


Changes in Sonic Annotator 0.4 since the previous release 0.3:

Build changes:

 - Sonic Annotator now requires the Vamp plugin SDK v2.1 or newer.

Front-end changes:

 - Frequency-domain plugins now use the PluginInputDomainAdapter's new
 ShiftData processing method, ensuring that the first block received
 by the plugin is the one centred on 0 rather than starting at 0.
 Unfortunately, the old behaviour omitted a timing compensation step,
 and this fix actually changes the results from some plugin processes:
 the previous behaviour was not always in line with the Vamp plugin
 specification.

 - More useful diagnostics are now available when a plugin fails to
 load or run.

CSV back-end (feature writer) changes:

 - The CSV writer now closes its output files properly when it
 finishes writing to each one, rather than leaving them all open
 until the end and risk running out of file descriptors.  This was
 already the behaviour of the RDF writer, the CSV one has just been
 updated to match it.
 - The behaviour of the CSV writer has also been changed to match
 that of the RDF writer in handling file write failures (it now
 continues processing only if --force is given).


Changes in Sonic Annotator 0.3 since the previous release 0.2:

Front-end changes:

 - A new --force front-end option (distinct from the --csv-force and
 --rdf-force back-end options) has been added, which makes Sonic
 Annotator continue to process subsequent audio files instead of
 exiting after an error, if multiple audio files have been specified.

RDF back-end (feature writer) changes:

 - The RDF writer now conforms more correctly with the Music Ontology
 in the way it relates signal, track, and track metadata.  Signals
 that have available metadata now gain a Track resource to associate
 that metadata with, instead of hanging it directly from the Signal
 resource; also the audio file now encodes the signal rather than the
 signal being available as the audio file (matching the domain
 specification of the Music Ontology).  Note that this new resource
 structure will not be properly read by versions of Sonic Visualiser
 prior to 1.6 (should importing the data into Sonic Visualiser be of
 interest to you).

 - The --rdf-signal-uri option has been removed and replaced with the
 more meaningful set of options --rdf-audiofile-uri, --rdf-track-uri,
 and --rdf-maker-uri.

 - A new --rdf-network option is available, to cause Sonic Annotator
 to try to retrieve RDF descriptions for plugins from the network
 where those descriptions are not available locally.  Use of this
 option is recommended, but it is not the default because of the
 possible performance implication (even though the results are mostly
 cached, there may be some network access involved).

 - The RDF writer now writes the computed_by property for signal
 features.

 - Plugin and output URIs in the resulting RDF are now percent-encoded.