CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

Unreleased

Changed

• migrated the project to a pyproject.toml implementation

Fixed

• eaf_builder replicating the subtree structure of recordings is not happening anymore. And individual files are not placed in individual subfolder anymore
• child-project overview now correctly displays the amount of audio
• audio duration in overview now works!

[0.1.2] 2024-03-14

Added

• add the derived annotation pipeline
• audio conversion has now a 'standard' conversion pipeline that will convert files to mono channel 16kHz pcm_s16le (no need to convert channels and then sampling rate and no need to know the options)
• add the simple_CTC metric to the list of available metrics

Changed

• the output of the CLI in the terminal is now handled by the logger module and not by print statements
• validating a dataset now results in warnings for broken symlinks and no errors anymore (#425)
• validation with recordings existing but for which mediainfo can't read the sample rate no longer fail but outputs a warning.
• recording_device_type in recordings.csv now accepts izyrec
• rework of the list of available metrics functions: now has a absolute value and per_hour decorator, and a peak_ option

Fixed

• zooniverse chunks upload was failing if the dataset column was missing in the csv
• eaf_builder with speaker_id NA no longer fails (#438)
• name of files in rttm that contain only digits in their name now work correctly with the filter ($457)
• periodic sampler having extra NA rows when using explode

[0.1.1] 2023-05-11

Added

• validation of the annotation index checks for annotation period outside of audio duration

Changed

• ignore_discarded is now the default behaviour of childproject projects, robustness was added to the discard column
• annotation index validation and annotation importation now checks for the range_offset being greater than duration and oiutputs an error when it is the case
• projects now check for unicity of the experiment column in children and recording csv, read() fails when not unique
• zooniverse uploads : uploads extra metadata to zooniverse (name of the audio clip, dataset it belongs to)
• zooniverse uploads : the subject_set id is stored in the chunk csv file, as the subject_set name (display name) is susceptible to change
• zooniverse uploads : the upload now handles SIGINT and SIGTERM signals to save progression of the upload to the csv before exiting (useful when a job needs to be interrupted
• allow once again get_within_time_range to take str arguments as times
• add arguments to choose the format of compute_ages project method

Fixed

• discard column in recordings.csv and children.csv now works properly
• metrics pipeline now checks the converted name for unicity even if a specific name was given
• rename set also renames the merged_from column
• rename set accepts subsets location without failing

[0.1.0] 2023-02-20

Added

• Windows automated tests (some functions were edited to be windows compatible). TO REMEMBER: type int in windows default to int32 instead of int64, can lead to big int turned into negative values
• Check calculated ages during corpus validation

Changed

• pandas version restricted to avoid errors of future releases , (1.1.0 (assert_frame_equal check_less_precise) to 1.5.0 (last checked version))
• no usage of sox command anymore, remove sox dependency
• merging annotations now sets the format to 'NA' instead of a blank value.

Fixed

• replace exit() with raise ValueError() to comply with Exception propagation (metrics pipeline)
• fixed ignore-errors in zooniverse upload_chunks
• fixed calculate_shift to correctly reshape to single channel

0.0.7 2022-09-14

Fixed

• missing column merged_from in annotations.csv does not fail anymore

0.0.6 2022-09-13

Added

• Start times can include seconds (e.g. 12:34:59) while still accepting the old format (This change will allow other columns to accept multiple formats easily).
• child-project --version command
• merge_sets in AnnotationManager method now accepts arguments [full_set_merge,skip_existing,recording_filter] to carry out partial merges
• child-project metrics added the --segments command to extract metrics from a dataFrame of segments
• metrics <voc_speaker> <lena_CTC> and <lena_CVC>

Changed

• metrics pipeline's options --from --to require a HH:MM:SS format now.
• merge-annotations command fails when the output_set already exists or if the required sets don't exist

Fixed

• eafbuilder attributes a default time-aligneable ling-type to created tiers to avoid random attribution that can lead to wrong behaviour and crashes
• 'imported_at' column in annotations.csv did not have a new correct format (in a set)
• metrics avg_cry_... avg_can_... and avg_non_can_... were not calculated correctly
• metrics lp_n lp_dur use lena columns in priority
• metrics lena-CTC and lena-CVC are added correctly and added to the output of the lena pipeline
• praat-parselmouth is now in the setup file so the dependency get installed automatically by conda

Dropped

• child-project process --split --split option dropped as there is no further need of reducing long audios (>15hs)

0.0.5 - 2022-07-25

Added

• --spectrogram option in the zooniverse extract-chunks pipeline to generate an image of a spectrogram that will help citizen-scientists with the classification on zooniverse.
• child-project compare-recordings command added to allow users to prin a divergence score. This will help identify audio files that are just duplicates of others (and possibily have different codecs/sampling rate/number of channels)
• --import-speech-from command added to the eaf builder to integrate previous annotations to the eaf file (e.g. VTC segments) to facilitate annotation process

Changed

• metrics pipeline, reworked to be more flexible. Performance hit with it.
  • Old pipelines still exist
  • new usage of --period option on every pipeline and for eveery metric.
  • Usage of a csv file to specify the list of metrics wanted
  • Ease of adding new metrics to the supported list
  • Outputs a yml parameter file that can be reused to compute the same metrics and keep a trace of what was run.
• changes to standard annotation value (addressee, vcm_type etc)

Fixed

• importation of empty file now correctly generates an empty converted file
• --period option correctly works with other units than recording_filename

Dropped

• Support for python 3.6

0.0.4 - 2022-02-02

Added

• Conversation sampler
• get_within_ranges function to retrieve all annotations that match the desired portions of the recordings
• --import-speech-from option for the EAF annotation builder to pre-fill annotations based on any previously imported set of annotations
• compute_ages function to compute the age of the subject child for each recording
• lena_speaker for the LENA its converter
• lena_speaker aggregated metrics for the LenaMetrics pipeline
• Improved AclewMetrics and LenaMetrics performance
• Improved error handling (dataframes sanity checks)

Changed

• More flexible high-volubility sampler

Fixed

• Fixed pipelines crashes in presence of NA values in recording_filename
• RandomVocalizationSampler crash fix

0.0.3 - 2021-10-06

Fixed

• Fixed exceptions thrown by child-project CLI

0.0.2 - 2021-09-29

Added

• CSV importer to register pre-exisiting CSV annotations into the index without performing any conversion
• Enable Zooniverse pipeline's retrieve-classifications to match classifications to the original chunks metadata
• get_within_time_range method to clip a list of annotations within a given time-range
• get_segments_timestamps method to calculate the onset and offset timestamps of each segment from an annotation
• --from-time/--to-time option for metrics extraction
• Time-unit aggregated metrics, supporting custom time periods.
• optional --recordings option to apply the audio processors to specific recordings only
• allow child-project validate to check custom recordings profiles and/or annotation sets
• --ignore-errors switch for Zooniverse pipeline's upload-chunks
• enforce_dtypes option for ChildProject in order to enforce the dtype of certain metadata columns (e.g. session_id, child_id) to their expected dtype

Fixed

• Fixed skip-existing argument of the basic audio processor
• Fixed a crash-bug that occured while extracting metrics from recordings with no FEM/MAL/CHI/OCH segment
• Made pyannote-agreement an optional dependency
• Added dependency constraints to fix some installation issues.

0.0.1 - 2021-07-14

Added

• First proper release of the package.