NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI, #3518 most downloaded on PyPI
A toolkit for reproducible evaluation, diagnostic, and error analysis of speaker diarization systems
Latest stable
4.2.0
Last release 1 months ago
16 Oct 2026
Release timing varies
gaps range from 8 days to 2.6 years
Most releases are documented
notes for 48 of 57 stable releases
Nothing withdrawn
no release was ever pulled
12 years old
59 releases · first in 2014
Drag across the ruler to compare, or tap a release to open it.
Python 3.11 or newer is now required. Python 3.10 is no longer supported.
python -O) is enabled.pip install "pyannote.metrics[transcription]".transcription dependenciesYour coding agent can read these notes before it upgrades. Set up the MCP server →
feat: add transcription metrics
fix: remove deprecated use of np.NaN
np.NaNPython < 3.10uvNothing published for this version
Nothing published for this version
fix: fix corner case for confidence interval
feat: add option to override existing "uri"
BREAKING: remove (buggy) support for parallel processing
setup: switch to pyannote.database 4.0+
feat: add DetectionCostFunction detection metric (@nryant)
feat: add DetectionPrecisionRecallFMeasure compound metric (@MarvinLvn)
feat: add support for evaluation of overlapped speech detection
chore: rewrite mapping and matching routines
- fix: avoid division by zero
BREAKING: drop support for all file formats but RTTM
Nothing published for this version
setup: switch to pyannote.core 2.0
feat: add compound segmentation metric SegmentationPurityCoverageFMeasure (@diego-fustes)
feat: add option to filter out target trials in "spotting" mode
fix: fix Diarization{Purity | Coverage} with empty references
Nothing published for this version
feat: add fixed vs. variable latency switch for LLSS
fix: add more safety checks to pyannote-metrics.py "spotting" mode
improve: set latency of missed detections to maximum possible value
feat: add LowLatencySpeakerSpotting metric
Nothing published for this version
feat: add "skip_overlap" option to not evaluate overlapping speech regions
feat: add method DiarizationPurityCoverageFMeasure.compute_metrics to get purity, coverage, and their F-measure (all at once)
feat: add new metric 'DiarizationPurityCoverageFMeasure'
setup: switch to pyannote.core 1.0
doc: update notebook to latest version
- doc: add Sphinx documentation
feat: better README and technical report
chore: rename SegmentationError to SegmentationErrorAnalysis
feat: add "parallel" option to not use multiprocessing
feat: add pyannote-metrics.py evaluation script
Nothing published for this version
Nothing published for this version
feat: support for multiprocessing
Nothing published for this version
feat: add pyannote.metrics.binary_classification module
fix: fix (greedy) diarization error rate
fix: fix default "xlim" in "plot_distributions"
feat: add "uem" support to diarization metrics
Nothing published for this version
feat: add plotting functions for binary classification tasks
setup: update to pyannote.core 0.7.2
setup: update to pyannote.core 0.6.6
feat: greedy diarization error rate
- feat: Python 3 support - feat: unit tests - wip: travis
- refactor: diarization metrics
Nothing published for this version
fix: identification error analysis matrix confusion
feat(error): identification regression analysis
feat(error): segmentation error analysis
feat(detection): add precision and recall
Nothing published for this version
feat(segmentation): add precision and recall
- first public version
One column per quarter.