pyannote-database

• chore: remove deprecated use of delim_whitespace

• chore: use importlib.metadata instead of pkg_resources

• BREAKING: switch to python 3.9+ (no support for 3.12 yet) - feat(registry): make registry iterable (yields protocol names) - fix(setup): fix typer` dependency

• fix: fix support for numeric values in custom protocols (e.g. channel: 0) - improve: do not complain about missing scope for meta protocols

• feat(registry): add pyannote.database.registry to load multiple YAML configuration file - BREAKING: turn pyannote.database.get_{database|protocol|protocols} function into registry methods - BREAKING: remove support for database plugins (via "pyannote.database.database" entrypoint) - feat(protocol): add generic segmentation protocol with "classes" metadata - feat(protocol): add support for label "scope" in speaker diarization protocols - improve(protocol): make ProtocolFile pickable - feat(loader): add support for LAB and STM files - feat(loader): add option to filter RTTM lines by "type" - fix(cli): fix corner case when annotation key is missing

• fix: fix corner case with missing uris in RTTM and UEM files - fix(CI): fix documentation deployment

• fix: fix support for meta verification protocol

• feat: make {RTTM|UEM}Loader usable as preprocessors - improve: remove the need for underscore prefix in database.yml - fix: fix ProtocolFile.__iter__ (dictionary changed size during iteration)

• fix: fix ProtocolFile.__iter__ (dictionary changed size during iteration)

• fix: make custom protocols pickable

• feat: automagically crop "annotated" to file "duration" when available - fix: fix a bug in SpeakerDiarizationProtocols - improve: avoid loading ProtocolFile (lazy) keys in "get_unique_identifier"

• feat: set "name" attribute in get_protocol - fix: display warning only when precomputed value is modified

• feat: add support for custom speaker verification protocols - feat: add pyannote.database.loader entrypoint - feat: add pyannote-database CLI - feat: add a few dataloaders (RTTM, UEM, CTM, MAP) - feat: add support for nested ProtocolFile - doc: major documentation update (README and docstrings) - BREAKING: custom protocols must define a "uri" section - BREAKING: remove support for "preprocessors" in Database constructor - BREAKING: remove support for progress bars

• BREAKING (feat): if a "database.yml" file exists in current working directory, it will be used even if PYANNOTE_DATABASE_CONFIG is set to another value. - feat: add support in FileFinder for paths relative to "database.yml" - BREAKING: rename "config_yml" option to "database_yml" in FileFinder - feat: add support in custom protocols for paths relative to "database.yml" (@PaulLerner) - BREAKING (feat): use "annotated" to crop "annotation" in custom protocols (@PaulLerner) - fix: add support for int-like protocol name in custom protocols (@PaulLerner)

• BREAKING: refactor {current | protocol}_file_iter - BREAKING: only rely on "uri" to decide if a ProtocolFile contains multiple files - BREAKING: deprecate FileFinder.current_file_iter in favor of ProtocolFile.files - BREAKING: deprecate FileFinder.protocol_file_iter in favor of Protocol.files - fix: fix support for lazy preprocessors in {Protocol | ProtocolFile}.files

• fix: fix infinite recursion in "ProtocolFile" lazy evaluation

• fix: add support for int-like database name in custom protocol

• fix: make ProtocolFile thread-safe

• feat: make preprocessors lazy - fix: pandas would convert a label to NaN ([@PaulLerner](https://github.com/PaulLerner)) - feat: setup continuous integration - setup: switch to pyannote.core 3.2

• fix: fix support for MDTM files in pyannote.database.custom.subset_iter ([#23](https://github.com/pyannote/pyannote-database/issues/23))

• feat: add LabelMapper preprocessor ([@MarvinLvn](https://github.com/MarvinLvn)) - chore: replace (deprecated) pandas.read_table with pandas.read_csv ([@V-assim](https://github.com/V-assim)) - chore: use YAML safe loader ([@V-assim](https://github.com/V-assim))

• setup: switch to pyannote.core 3.0 - feat: add RTTMLoader preprocessor

• feat: add support for "domain" key in generic protocol

• BREAKING: change location and format of pyannote.database configuration file - feat: add support for PYANNOTE_DATABASE_CONFIG environment variable

• feat: add support for [file-based speaker diarization protocols](https://github.com/pyannote/pyannote-database/tree/develop#generic-speaker-diarization-protocols) - setup: switch to pyannote.core 2.1

• fix: fix Collection.files when progress=True

• fix: skip files with no "uri" entry in FileFinder.protocol_file_iter

• fix: fix broken SpeakerVerificationProtocol

• fix: fix support for string preprocessors

• BREAKING: simplify SpeakerVerificationProtocol with {subset}_trial methods

• feat: add raw collection protocol

• fix: fix regression introduced in 1.3.1

• fix: fix bug in FileFinder.protocol_file_iter with empty iterators

• feat: add extra_keys parameter to {protocol | current}_file_iter

• setup: drop support for Python 2 - feat: add protocol_file_iter and current_file_iter to FileFinder - feat: add get_label_identifier utility function - fix: fix "get_unique_identifier" when "database" or "channel" is None

• feat: add speaker identification protocol - feat: add speaker verification protocols - feat: add support for list of uris in FileFinder

• feat: add support for "meta" protocols - feat: add speaker spotting protocol - setup: switch to pyannote.core 1.1

• feat: add utility functions at package root - doc: improve documentation - doc: add link to pyannote-db-template repository

• fix: fix a bug with string template preprocessors - doc: improve documentation

• feat: add 'get_protocol' helper function

• feat: add support for validation on training set to speaker recognition protocols - feat: add 'get_annotated' helper function

• fix: fix bug in FileFinder

• improve: change signature of preprocessor.__call__

• feat: add "get_unique_identifier" utility function

• fix: fix progress bar support

• feat: add progress bar support

• fix: add 'yield_name' parameter to speaker recognition generators

• feat: add speaker recognition protocol

• feat: add FileFinder utility class - fix: fix SpeakerDiarizationProtocol.stats()

• BREAKING: replace 'medium_template' by (more generic) 'preprocessors'