BREAKING: switch to uv, namespace package, Python 3.10+, remove old CLI

This commit is contained in:
Hervé BREDIN authored and GitHub committed 2025-02-12 13:07:16 +01:00
1 parent d9389defa7
commit 93ad8b90ca
155 files changed
+15125 -3529

No files matched your search

-2
View File
@@ -1,4 +1,2 @@
pyannote/audio/_version.py export-subst
notebooks/* linguist-documentation
tutorials/* linguist-documentation
versioneer.py linguist-vendored
-3
View File
@@ -1,3 +0,0 @@
# These are supported funding model platforms
github: [hbredin]
+16 -21
View File
@@ -7,32 +7,27 @@ on:
jobs:
build-and-deploy:
runs-on: ubuntu-latest
strategy:
max-parallel: 4
matrix:
python-version: ["3.9"]
steps:
- uses: actions/checkout@v1
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v1
- uses: actions/checkout@v4
with:
python-version: ${{ matrix.python-version }}
- name: Install
run: |
python -m pip install --upgrade pip
pip install .
pip install -r doc/requirements.txt
persist-credentials: false
fetch-depth: 0
- name: Install uv
uses: astral-sh/setup-uv@v5
with:
enable-cache: true
cache-dependency-glob: uv.lock
- name: Install the project
run: uv sync --extra doc
- name: Build documentation
run: |
make --directory=doc html
touch ./doc/build/html/.nojekyll
- name: Deploy
env:
ACTIONS_DEPLOY_KEY: ${{ secrets.ACTIONS_DEPLOY_KEY }}
PUBLISH_BRANCH: gh-pages
PUBLISH_DIR: ./doc/build/html
SCRIPT_MODE: true
run: |
wget https://raw.githubusercontent.com/peaceiris/actions-gh-pages/v2/entrypoint.sh
bash ./entrypoint.sh
uses: peaceiris/actions-gh-pages@v3
with:
github_token: ${{ secrets.GITHUB_TOKEN }}
publish_dir: ./doc/build/html
-27
View File
@@ -1,27 +0,0 @@
name: PyPI
on:
push:
tags:
- '*'
jobs:
deploy:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v1
- name: Set up Python
uses: actions/setup-python@v1
with:
python-version: '3.x'
- name: Install dependencies
run: |
python -m pip install --upgrade pip
pip install setuptools wheel twine
- name: Build and publish
env:
TWINE_USERNAME: ${{ secrets.PYPI_USERNAME }}
TWINE_PASSWORD: ${{ secrets.PYPI_PASSWORD }}
run: |
python setup.py sdist bdist_wheel
twine upload dist/*
+125
View File
@@ -0,0 +1,125 @@
name: Publish Python 🐍 distribution 📦 to PyPI and TestPyPI
on: push
jobs:
build:
name: Build distribution 📦
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
with:
persist-credentials: false
fetch-depth: 0
- name: Install uv
uses: astral-sh/setup-uv@v5
with:
enable-cache: true
cache-dependency-glob: uv.lock
- name: Set up Python
uses: actions/setup-python@v5
with:
python-version-file: ".python-version"
- name: Build
run: uv build
- name: Store the distribution packages
uses: actions/upload-artifact@v4
with:
name: python-package-distributions
path: dist/
publish-to-pypi:
name: >-
Publish Python 🐍 distribution 📦 to PyPI
if: startsWith(github.ref, 'refs/tags/') # only publish to PyPI on tag pushes
needs:
- build
runs-on: ubuntu-latest
environment:
name: pypi
permissions:
id-token: write
steps:
- name: Download all the dists
uses: actions/download-artifact@v4
with:
name: python-package-distributions
path: dist/
- name: Install uv
uses: astral-sh/setup-uv@v5
with:
enable-cache: true
cache-dependency-glob: uv.lock
- name: Publish distribution 📦 to PyPI
run: uv publish --trusted-publishing always --publish-url https://upload.pypi.org/legacy/
github-release:
name: >-
Sign the Python 🐍 distribution 📦 with Sigstore
and upload them to GitHub Release
needs:
- publish-to-pypi
runs-on: ubuntu-latest
permissions:
contents: write # IMPORTANT: mandatory for making GitHub Releases
id-token: write # IMPORTANT: mandatory for sigstore
steps:
- name: Download all the dists
uses: actions/download-artifact@v4
with:
name: python-package-distributions
path: dist/
- name: Sign the dists with Sigstore
uses: sigstore/gh-action-sigstore-python@v3.0.0
with:
inputs: >-
./dist/*.tar.gz
./dist/*.whl
- name: Create GitHub Release
env:
GITHUB_TOKEN: ${{ github.token }}
run: >-
gh release create
"$GITHUB_REF_NAME"
--repo "$GITHUB_REPOSITORY"
--notes ""
- name: Upload artifact signatures to GitHub Release
env:
GITHUB_TOKEN: ${{ github.token }}
# Upload to GitHub Release using the `gh` CLI.
# `dist/` contains the built packages, and the
# sigstore-produced signatures and certificates.
run: >-
gh release upload
"$GITHUB_REF_NAME" dist/**
--repo "$GITHUB_REPOSITORY"
# publish-to-testpypi:
# name: Publish Python 🐍 distribution 📦 to TestPyPI
# needs:
# - build
# runs-on: ubuntu-latest
#
# environment:
# name: testpypi
#
# permissions:
# id-token: write # IMPORTANT: mandatory for trusted publishing
#
# steps:
# - name: Download all the dists
# uses: actions/download-artifact@v4
# with:
# name: python-package-distributions
# path: dist/
# - name: Install uv
# uses: astral-sh/setup-uv@v5
# with:
# enable-cache: true
# cache-dependency-glob: uv.lock
# - name: Publish distribution 📦 to PyPI
# run: uv publish --trusted-publishing always --publish-url https://test.pypi.org/legacy/
+25 -25
View File
@@ -1,33 +1,33 @@
name: Tests
name: Test
on:
push:
branches: [develop]
pull_request:
branches: [develop]
branches:
- develop
push:
branches:
- develop
- master
- release/*
jobs:
build:
timeout-minutes: 20
runs-on: ${{ matrix.os }}
test:
name: Test
runs-on: ubuntu-latest
strategy:
matrix:
os: [ubuntu-latest]
python-version: ["3.9", "3.10", "3.11"]
python-version:
- "3.10"
- "3.11"
- "3.12"
env:
UV_PYTHON: ${{ matrix.python-version }}
steps:
- uses: actions/checkout@v2
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Install libsndfile
if: matrix.os == 'ubuntu-latest'
run: |
sudo apt-get update
sudo apt-get install libsndfile1
- name: Install pyannote.audio
run: |
pip install -e .[dev,testing]
- name: Test with pytest
run: |
pytest -k "not test_cli.py"
- uses: actions/checkout@v4
- name: Install uv
uses: astral-sh/setup-uv@v5
- name: Run tests
run: uv run --extra test pytest
-33
View File
@@ -1,33 +0,0 @@
name: CLI tests
on:
push:
branches: [develop]
pull_request:
branches: [develop]
jobs:
build:
timeout-minutes: 20
runs-on: ${{ matrix.os }}
strategy:
matrix:
os: [ubuntu-latest]
python-version: ["3.10"]
steps:
- uses: actions/checkout@v2
- name: Set up Python ${{ matrix.python-version }}
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Install libsndfile
if: matrix.os == 'ubuntu-latest'
run: |
sudo apt-get update
sudo apt-get install libsndfile1
- name: Install pyannote.audio
run: |
pip install -e .[dev,testing,cli]
- name: Test with pytest
run: |
pytest tests/test_cli.py
+1
View File
@@ -0,0 +1 @@
3.10
+5 -1
View File
@@ -1,4 +1,4 @@
# Changelog
# CHANGELOG
## develop
@@ -18,6 +18,8 @@ Clipping and speaker/source alignment issues in speech separation pipeline have
### Breaking changes
- BREAKING(setup): drop support to `Python` < 3.10
- BREAKING(chore): switch to native namespace package
- BREAKING(hub): rename `use_auth_token` to `token`
- BREAKING(cache): rely on `huggingface_hub` caching directory (`PYANNOTE_CACHE` is no longer used)
- BREAKING(inference): `Inference` now only supports already instantiated models
@@ -25,6 +27,7 @@ Clipping and speaker/source alignment issues in speech separation pipeline have
- BREAKING(task): drop support for `warm_up` option in `SpeakerDiarization` task
- BREAKING(task): drop support for `weigh_by_cardinality` option in `SpeakerDiarization` task
- BREAKING(task): drop support for `vad_loss` option in `SpeakerDiarization` task
- BREAKING(cli): remove deprecated `pyannote-audio-train` CLI
### New features
@@ -42,6 +45,7 @@ Clipping and speaker/source alignment issues in speech separation pipeline have
- improve(model): improve WavLM (un)freezing support for `SSeRiouSS` architecture ([@clement-pages](https://github.com/clement-pages/))
- improve(task): improve `SpeakerDiarization` training with manual optimization ([@clement-pages](https://github.com/clement-pages/))
- improve(setup): switch to `uv`
### Fixes
+4 -4
View File
@@ -1,6 +1,6 @@
recursive-include pyannote *.py
recursive-include pyannote *.yaml
recursive-include pyannote *.wav
recursive-include pyannote *.rttm
recursive-include src/pyannote *.py
recursive-include src/pyannote *.yaml
recursive-include src/pyannote *.wav
recursive-include src/pyannote *.rttm
global-exclude *.pyc
global-exclude __pycache__
-5
View File
@@ -1,5 +0,0 @@
coverage:
status:
patch:
default:
enabled: false
+20
View File
@@ -0,0 +1,20 @@
# Minimal makefile for Sphinx documentation
#
# You can set these variables from the command line.
SPHINXOPTS =
SPHINXBUILD = sphinx-build
SPHINXPROJ = pyannoteaudio
SOURCEDIR = source
BUILDDIR = build
# Put it first so that "make" without argument is like "make help".
help:
uv run --extra doc $(SPHINXBUILD) -M help "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
.PHONY: help Makefile
# Catch-all target: route all unknown targets to Sphinx using the new
# "make mode" option. $(O) is meant as a shortcut for $(SPHINXOPTS).
%: Makefile
uv run --extra doc $(SPHINXBUILD) -M $@ "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
-63
View File
@@ -1,63 +0,0 @@
"""
This script will generate the rst docs for the api
"""
import os
from os import path
bp = breakpoint
def capitalise(s):
news = ""
for word in s.split("_"):
news += word.capitalize()
return news
def process_dir(level, p):
md = ""
basename = path.basename(p)
title = capitalise(basename)
md += f"{'#'*level} {title}\n\n"
subdirs = os.listdir(p)
for f in subdirs:
m = path.join(subdir, f)
if path.isdir(m):
md += process_dir(level + 1, path.join(p, f))
else:
if "__" in f:
continue
module = m[3:].replace("/", ".")[:-3]
md += f"""
```eval_rst
.. automodule:: {module}
:members:
```
"""
return md
DIR = "../pyannote/audio"
for module in os.listdir(DIR):
# Each folder will become and rst file
# Each file/folder will have a # prepended to it
# Recursively we will add another # each level
# Initialise Markdown
md = ""
subdir = path.join(DIR, module)
# Skip if not directory
if not path.isdir(subdir) or "__" in module:
continue
md += process_dir(1, subdir)
with open(f"./source/api/{module}.md", "w") as f:
f.write(md)
-4
View File
@@ -1,4 +0,0 @@
ipython==8.10.0
recommonmark
Sphinx==3.0.4
sphinx_rtd_theme==0.4.3
+196
View File
@@ -0,0 +1,196 @@
# -*- coding: utf-8 -*-
#
# pyannote.audio documentation build configuration file, created by
# sphinx-quickstart on Tue Jan 24 15:45:55 2017.
#
# This file is execfile()d with the current directory set to its
# containing dir.
#
# Note that not all possible configuration values are present in this
# autogenerated file.
#
# All configuration values have a default; values that are commented out
# serve to show the default.
# If extensions (or modules to document with autodoc) are in another directory,
# add these directories to sys.path here. If the directory is relative to the
# documentation root, use os.path.abspath to make it absolute, like shown here.
#
# allow pyannote.audio import
import os
import sys
sys.path.insert(0, os.path.abspath("../.."))
# -- General configuration ------------------------------------------------
# If your documentation needs a minimal Sphinx version, state it here.
#
# needs_sphinx = '1.0'
# Add any Sphinx extension module names here, as strings. They can be
# extensions coming with Sphinx (named 'sphinx.ext.*') or your custom
# ones.
extensions = [
"sphinx.ext.autodoc",
"sphinx.ext.napoleon",
"sphinx.ext.intersphinx",
"sphinx.ext.todo",
"sphinx.ext.coverage",
"sphinx.ext.mathjax",
"sphinx.ext.viewcode",
"sphinx.ext.githubpages",
]
# Napoleon settings
napoleon_google_docstring = True
napoleon_numpy_docstring = True
napoleon_include_init_with_doc = False
napoleon_include_private_with_doc = False
napoleon_include_special_with_doc = False
napoleon_use_admonition_for_examples = False
napoleon_use_admonition_for_notes = False
napoleon_use_admonition_for_references = False
napoleon_use_ivar = False
napoleon_use_param = True
napoleon_use_rtype = True
napoleon_use_keyword = True
# Add any paths that contain templates here, relative to this directory.
templates_path = ["_templates"]
# The suffix(es) of source filenames.
# You can specify multiple suffix as a list of string:
#
# source_suffix = ['.rst', '.md']
source_suffix = {".rst": "restructuredtext"}
# The master toctree document.
master_doc = "index"
# General information about the project.
project = "pyannote.audio"
copyright = "2017, CNRS"
author = "Hervé Bredin"
# The version info for the project you're documenting, acts as replacement for
# |version| and |release|, also used in various other places throughout the
# built documents.
import pyannote.audio # noqa
# The short X.Y version.
version = pyannote.audio.__version__.split("+")[0]
# The full version, including alpha/beta/rc tags.
release = pyannote.audio.__version__
# The language for content autogenerated by Sphinx. Refer to documentation
# for a list of supported languages.
#
# This is also used if you do content translation via gettext catalogs.
# Usually you set "language" from the command line for these cases.
language = "en"
# List of patterns, relative to source directory, that match files and
# directories to ignore when looking for source files.
# This patterns also effect to html_static_path and html_extra_path
exclude_patterns = []
# The name of the Pygments (syntax highlighting) style to use.
pygments_style = "sphinx"
# If true, `todo` and `todoList` produce output, else they produce nothing.
todo_include_todos = True
# -- Options for HTML output ----------------------------------------------
# The theme to use for HTML and HTML Help pages. See the documentation for
# a list of builtin themes.
#
html_theme = "sphinx_rtd_theme"
# Theme options are theme-specific and customize the look and feel of a theme
# further. For a list of options available for each theme, see the
# documentation.
#
# html_theme_options = {}
# Add any paths that contain custom static files (such as style sheets) here,
# relative to this directory. They are copied after the builtin static files,
# so a file named "default.css" will overwrite the builtin "default.css".
html_static_path = ["_static"]
# -- Options for HTMLHelp output ------------------------------------------
# Output file base name for HTML help builder.
htmlhelp_basename = "pyannoteaudiodoc"
# -- Options for LaTeX output ---------------------------------------------
latex_elements = {
# The paper size ('letterpaper' or 'a4paper').
#
# 'papersize': 'letterpaper',
# The font size ('10pt', '11pt' or '12pt').
#
# 'pointsize': '10pt',
# Additional stuff for the LaTeX preamble.
#
# 'preamble': '',
# Latex figure (float) alignment
#
# 'figure_align': 'htbp',
}
# Grouping the document tree into LaTeX files. List of tuples
# (source start file, target name, title,
# author, documentclass [howto, manual, or own class]).
latex_documents = [
(
master_doc,
"pyannoteaudio.tex",
"pyannote.audio Documentation",
"Hervé Bredin",
"manual",
),
]
# -- Options for manual page output ---------------------------------------
# One entry per manual page. List of tuples
# (source start file, name, description, authors, manual section).
man_pages = [(master_doc, "pyannoteaudio", "pyannote.audio Documentation", [author], 1)]
# -- Options for Texinfo output -------------------------------------------
# Grouping the document tree into Texinfo files. List of tuples
# (source start file, target name, title, author,
# dir menu entry, description, category)
texinfo_documents = [
(
master_doc,
"pyannoteaudio",
"pyannote.audio Documentation",
author,
"pyannoteaudio",
"One line description of project.",
"Miscellaneous",
),
]
# Example configuration for intersphinx: refer to the Python standard library.
intersphinx_mapping = {
"python": ("https://docs.python.org/", None),
"pyannote.core": ("https://pyannote.github.io/pyannote-core", None),
"pyannote.database": ("https://pyannote.github.io/pyannote-database", None),
"pyannote.metrics": ("https://pyannote.github.io/pyannote-metrics", None),
"pyannote.pipeline": ("https://pyannote.github.io/pyannote-pipeline", None),
}
+2 -17
View File
@@ -1,21 +1,6 @@
##############
pyannote.audio
##############
`pyannote.audio` is an open-source Python library that provides neural building blocks for speaker diarization.
Installation
============
::
$ conda create -n pyannote python=3.10
$ conda activate pyannote
$ pip install pyannote.audio
API documentation
=================
.. toctree::
:maxdepth: 2
`pyannote.audio` is an open-source toolkit for state-of-the-art speaker diarization.
-10
View File
@@ -1,10 +0,0 @@
name: pyannote-audio
channels:
- defaults
- conda-forge
dependencies:
- python==3.8.5
- libsndfile==1.0.28
- pip>=20.2
- pip:
- -r requirements.txt
+1 -1
View File
@@ -29,7 +29,7 @@
"source": [
"from pyannote.audio.models.segmentation.debug import SimpleSegmentationModel\n",
"from pyannote.audio.models.embedding.debug import SimpleEmbeddingModel\n",
"import pytorch_lightning as pl"
"import lightning.pytorch as pl"
]
},
{
-23
View File
@@ -1,23 +0,0 @@
# MIT License
#
# Copyright (c) 2020 CNRS
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
__import__("pkg_resources").declare_namespace(__name__)
-27
View File
@@ -1,27 +0,0 @@
# MIT License
#
# Copyright (c) 2020-2021 CNRS
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
from .pretrained import pretrained
__all__ = [
"pretrained",
]
@@ -1,31 +0,0 @@
# @package _group_
run:
dir: ${protocol}/${now:%Y-%m-%dT%H:%M:%S.%fZ}
sweep:
dir: ${protocol}/${now:%Y-%m-%dT%H:%M:%S.%fZ}
subdir: ${hydra.job.num}
output_subdir: ""
help:
app_name: pyannote-audio-train
# Help header, customize to describe your app to your users
header: == ${hydra.help.app_name} ==
footer: |-
Powered by Hydra (https://hydra.cc)
Use --hydra-help to view Hydra specific help
template: |-
${hydra.help.header}
pyannote-audio-train protocol={protocol_name}
task={task} task.param=...
model={model} model.param=...
optimizer={optimizer} optimizer.param=...
scheduler={scheduler} scheduler.param=...
${hydra.help.footer}
-101
View File
@@ -1,101 +0,0 @@
# MIT License
#
# Copyright (c) 2022- CNRS
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
from typing import Optional
import hydra
from omegaconf import DictConfig
from pyannote.database import FileFinder, ProtocolFile, registry
from rich.progress import Progress
from pyannote.audio import Inference, Model
from pyannote.audio.pipelines.utils import get_devices
from pyannote.audio.utils.metric import DiscreteDiarizationErrorRate
from pyannote.audio.utils.signal import binarize
@hydra.main(config_path="evaluate_config", config_name="config")
def evaluate(cfg: DictConfig) -> Optional[float]:
# load pretrained model
(device,) = get_devices(needs=1)
model = Model.from_pretrained(cfg.model, device=device)
# load databases into registry if it was specified
if "registry" in cfg:
for database_yml in cfg.registry.split(","):
registry.load_database(database_yml)
# load evaluation files
protocol = registry.get_protocol(
cfg.protocol, preprocessors={"audio": FileFinder()}
)
files = list(getattr(protocol, cfg.subset)())
# load evaluation metric
metric = DiscreteDiarizationErrorRate()
with Progress() as progress:
main_task = progress.add_task(protocol.name, total=len(files))
file_task = progress.add_task("Processing", total=1.0)
def progress_hook(completed: Optional[int] = None, total: Optional[int] = None):
progress.update(file_task, completed=completed / total)
inference = Inference(model, device=device)
warm_up = cfg.warm_up / inference.duration
def hypothesis(file: ProtocolFile):
return Inference.trim(
binarize(inference(file, hook=progress_hook)),
warm_up=(warm_up, warm_up),
)
for file in files:
progress.update(file_task, description=file["uri"])
reference = file["annotation"]
uem = file["annotated"]
_ = metric(reference, hypothesis(file), uem=uem)
progress.advance(main_task)
report = metric.report(display=False)
with open("report.txt", "w") as f:
f.write(f"# Model: {cfg.model}\n")
f.write(f"# Protocol: {protocol.name}\n")
f.write(f"# Subset: {cfg.subset}\n")
f.write("\n")
report = report.to_string(
index=True,
sparsify=False,
justify="right",
float_format=lambda f: "{0:.2f}".format(f),
)
f.write(f"{report}")
if __name__ == "__main__":
evaluate()
@@ -1,21 +0,0 @@
# MIT License
#
# Copyright (c) 2020-2021 CNRS
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
@@ -1,7 +0,0 @@
model: ???
protocol: ???
warm_up: 0.0
subset: test
defaults:
- hydra: default
@@ -1,31 +0,0 @@
# @package _group_
run:
dir: ${protocol}/${now:%Y-%m-%dT%H:%M:%S.%fZ}
sweep:
dir: ${protocol}/${now:%Y-%m-%dT%H:%M:%S.%fZ}
subdir: ${hydra.job.num}
output_subdir: ""
help:
app_name: pyannote-audio-eval
# Help header, customize to describe your app to your users
header: == ${hydra.help.app_name} ==
footer: |-
Powered by Hydra (https://hydra.cc)
Use --hydra-help to view Hydra specific help
template: |-
${hydra.help.header}
pyannote-audio-eval registry={path_to_database.yml}
protocol={protocol_name}
subset={test | development | train}
model={path_to_pretrained_model}
warm_up={warm_up_duration_in_seconds}
${hydra.help.footer}
@@ -1,66 +0,0 @@
# MIT License
#
# Copyright (c) 2022 CNRS
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
from typing import Optional
from torch.optim import Optimizer
from torch.optim.lr_scheduler import (
CosineAnnealingWarmRestarts as _CosineAnnealingWarmRestarts,
)
def CosineAnnealingWarmRestarts(
optimizer: Optimizer,
min_lr: float = 1e-8,
max_lr: float = 1e-3,
patience: int = 1,
num_batches_per_epoch: Optional[int] = None,
**kwargs,
):
"""Wrapper around CosineAnnealingWarmRestarts
Parameters
----------
optimizer : Optimizer
Optimizer
min_lr : float, optional
Defaults to 1e-8.
max_lr : float, optional
Defaults to 1e-3
patience : int, optional
Number of epochs per cycle. Defaults to 1.
num_batches_per_epoch : int, optional
Number of batches per epoch.
"""
# initialize optimizer lr to max_lr
for g in optimizer.param_groups:
g["lr"] = max_lr
num_steps = patience * num_batches_per_epoch
return {
"scheduler": _CosineAnnealingWarmRestarts(
optimizer, num_steps, eta_min=min_lr, T_mult=2
),
"interval": "step",
}
@@ -1,68 +0,0 @@
# MIT License
#
# Copyright (c) 2021 CNRS
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
from typing import Optional
from torch.optim import Optimizer
from torch.optim.lr_scheduler import CyclicLR as _CyclicLR
def CyclicLR(
optimizer: Optimizer,
min_lr: float = 1e-8,
max_lr: float = 1e-3,
mode: str = "triangular2",
patience: int = 50,
num_batches_per_epoch: Optional[int] = None,
**kwargs,
):
"""Wrapper around CyclicLR learning rate scheduler
Parameters
----------
optimizer : Optimizer
Optimizer
min_lr : float, optional
Defaults to 1e-8.
max_lr : float, optional
Defaults to 1e-3
patience : int, optional
Number of epochs per cycle. Defaults to 50.
num_batches_per_epoch : int, optional
Number of batches per epoch.
mode : {"triangular", "triangular2"}, optional
Defaults to "triangular2".
"""
step_size_up = int(0.5 * patience * num_batches_per_epoch)
return {
"scheduler": _CyclicLR(
optimizer,
base_lr=min_lr,
max_lr=max_lr,
step_size_up=step_size_up,
mode=mode,
cycle_momentum=False,
),
"interval": "step",
}
@@ -1,81 +0,0 @@
# MIT License
#
# Copyright (c) 2021 CNRS
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
from typing import Optional, Text
from torch.optim import Optimizer
from torch.optim.lr_scheduler import ReduceLROnPlateau as _ReduceLROnPlateau
def ReduceLROnPlateau(
optimizer: Optimizer,
monitor: Optional[Text] = None,
direction: Optional[Text] = "min",
min_lr: float = 1e-8,
max_lr: float = 1e-3,
factor: float = 0.5,
patience: int = 50,
**kwargs,
):
"""Wrapper around ReduceLROnPlateau learning rate scheduler
Parameters
----------
optimizer : Optimizer
Optimizer
min_lr : float, optional
Defaults to 1e-8.
max_lr : float, optional
Defaults to 1e-3
factor : float, optional
Defaults to 0.5
patience : int, optional
Wait that many epochs with no improvement before reducing the learning rate.
Defaults to 50.
monitor : str, optional
Value to monitor
direction : {"min", "max"}, optional
"min" (resp. "max") means smaller (resp. larger) is better.
"""
# initialize optimizer lr to max_lr
for g in optimizer.param_groups:
g["lr"] = max_lr
return {
"scheduler": _ReduceLROnPlateau(
optimizer,
mode=direction,
factor=factor,
patience=patience,
threshold=0.0001,
threshold_mode="rel",
cooldown=0,
min_lr=min_lr,
eps=1e-08,
verbose=False,
),
"interval": "epoch",
"monitor": monitor,
"strict": True,
}
@@ -1,28 +0,0 @@
# MIT License
#
# Copyright (c) 2021 CNRS
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
from .CosineAnnealingWarmRestarts import CosineAnnealingWarmRestarts
from .CyclicLR import CyclicLR
from .ReduceLROnPlateau import ReduceLROnPlateau
__all__ = ["ReduceLROnPlateau", "CyclicLR", "CosineAnnealingWarmRestarts"]
-29
View File
@@ -1,29 +0,0 @@
# MIT License
#
# Copyright (c) 2020-2021 CNRS
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
from typing import Text
from pyannote.audio import Model
def pretrained(checkpoint: Text):
return Model.from_pretrained(checkpoint, map_location=lambda storage, loc: storage)
-171
View File
@@ -1,171 +0,0 @@
# MIT License
#
# Copyright (c) 2020-2022 CNRS
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
import os
from types import MethodType
from typing import Optional
import hydra
from hydra.utils import instantiate
from lightning.pytorch import seed_everything
from omegaconf import DictConfig, OmegaConf
# from pyannote.audio.core.callback import GraduallyUnfreeze
from pyannote.database import FileFinder, registry
from pytorch_lightning.callbacks import (
EarlyStopping,
LearningRateMonitor,
ModelCheckpoint,
RichProgressBar,
)
from pytorch_lightning.loggers import TensorBoardLogger
from torch_audiomentations.utils.config import from_dict as get_augmentation
from pyannote.audio.core.io import get_torchaudio_info
@hydra.main(config_path="train_config", config_name="config")
def train(cfg: DictConfig) -> Optional[float]:
# make sure to set the random seed before the instantiation of Trainer
# so that each model initializes with the same weights when using DDP.
seed = int(os.environ.get("PL_GLOBAL_SEED", "0"))
seed_everything(seed=seed)
# load databases into registry
if "registry" in cfg:
for database_yml in cfg.registry.split(","):
registry.load_database(database_yml)
# instantiate training protocol with optional preprocessors
preprocessors = {"audio": FileFinder(), "torchaudio.info": get_torchaudio_info}
if "preprocessor" in cfg:
preprocessor = instantiate(cfg.preprocessor)
preprocessors[preprocessor.preprocessed_key] = preprocessor
protocol = registry.get_protocol(cfg.protocol, preprocessors=preprocessors)
# instantiate data augmentation
augmentation = (
get_augmentation(OmegaConf.to_container(cfg.augmentation))
if "augmentation" in cfg
else None
)
if augmentation is not None:
augmentation.output_type = "dict"
# instantiate task and validation metric
task = instantiate(cfg.task, protocol, augmentation=augmentation)
# instantiate model
fine_tuning = cfg.model["_target_"] == "pyannote.audio.cli.pretrained"
model = instantiate(cfg.model, task=task)
model.prepare_data()
model.setup()
# validation metric to monitor (and its direction: min or max)
monitor, direction = task.val_monitor
# number of batches in one epoch
num_batches_per_epoch = model.task.train__len__() // model.task.batch_size
# configure optimizer and scheduler
def configure_optimizers(self):
optimizer = instantiate(cfg.optimizer, self.parameters())
lr_scheduler = instantiate(
cfg.scheduler,
optimizer,
monitor=monitor,
direction=direction,
num_batches_per_epoch=num_batches_per_epoch,
)
return {"optimizer": optimizer, "lr_scheduler": lr_scheduler}
model.configure_optimizers = MethodType(configure_optimizers, model)
# avoid creating big log files
callbacks = [
RichProgressBar(refresh_rate=20, leave=True),
LearningRateMonitor(),
]
if fine_tuning:
# TODO: configure layer freezing
# TODO: for fine-tuning and/or transfer learning, we start by fitting
# TODO: task-dependent layers and gradully unfreeze more layers
# TODO: callbacks.append(GraduallyUnfreeze(epochs_per_stage=1))
pass
checkpoint = ModelCheckpoint(
monitor=monitor,
mode=direction,
save_top_k=None if monitor is None else 10,
every_n_epochs=1,
save_last=True,
save_weights_only=False,
dirpath=".",
filename="{epoch}" if monitor is None else f"{{epoch}}-{{{monitor}:.6f}}",
verbose=False,
)
callbacks.append(checkpoint)
if monitor is not None:
early_stopping = EarlyStopping(
monitor=monitor,
mode=direction,
min_delta=0.0,
patience=100,
strict=True,
verbose=False,
check_finite=True,
)
callbacks.append(early_stopping)
# instantiate logger
logger = TensorBoardLogger(".", name="", version="", log_graph=False)
# instantiate trainer
trainer = instantiate(cfg.trainer, callbacks=callbacks, logger=logger)
# in case of fine-tuning, validate the initial model to make sure
# that we actually improve over the initial performance
if fine_tuning:
trainer.validate(model)
# train the model
trainer.fit(model)
# save paths to best models
checkpoint.to_yaml()
# return the best validation score
# this can be used for hyper-parameter optimization with Hydra sweepers
# this can only be done if the trainer is not in fast dev run mode, as
# checkpointing is disabled in this mode
if monitor is not None and not trainer.fast_dev_run:
best_monitor = float(checkpoint.best_model_score)
if direction == "min":
return best_monitor
else:
return -best_monitor
if __name__ == "__main__":
train()
@@ -1,21 +0,0 @@
# MIT License
#
# Copyright (c) 2020-2021 CNRS
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
@@ -1,9 +0,0 @@
protocol: ???
defaults:
- task: SpeakerDiarization
- model: PyanNet
- optimizer: Adam
- scheduler: CosineAnnealingWarmRestarts
- trainer: default
- hydra: default
@@ -1,31 +0,0 @@
# @package _group_
run:
dir: ${protocol}/${now:%Y-%m-%dT%H:%M:%S.%fZ}
sweep:
dir: ${protocol}/${now:%Y-%m-%dT%H:%M:%S.%fZ}
subdir: ${hydra.job.num}
output_subdir: ""
help:
app_name: pyannote-audio-train
# Help header, customize to describe your app to your users
header: == ${hydra.help.app_name} ==
footer: |-
Powered by Hydra (https://hydra.cc)
Use --hydra-help to view Hydra specific help
template: |-
${hydra.help.header}
pyannote-audio-train protocol={protocol_name}
+task={task} task.param=...
+model={model} model.param=...
optimizer={optimizer} optimizer.param=...
scheduler={scheduler} scheduler.param=...
${hydra.help.footer}
@@ -1,2 +0,0 @@
# @package _group_
_target_: pyannote.audio.models.embedding.debug.SimpleEmbeddingModel
@@ -1,2 +0,0 @@
# @package _group_
_target_: pyannote.audio.models.segmentation.debug.SimpleSegmentationModel
@@ -1,3 +0,0 @@
# @package _group_
_target_: pyannote.audio.cli.pretrained
checkpoint: ???
@@ -1,13 +0,0 @@
# @package _group_
_target_: pyannote.audio.models.segmentation.PyanNet
sincnet:
stride: 10
lstm:
hidden_size: 128
num_layers: 2
bidirectional: true
monolithic: true
dropout: 0.5
linear:
hidden_size: 128
num_layers: 2
@@ -1,13 +0,0 @@
# @package _group_
_target_: pyannote.audio.models.segmentation.SSeRiouSS
wav2vec: WAVLM_BASE
wav2vec_layer: -1
lstm:
hidden_size: 128
num_layers: 4
bidirectional: true
monolithic: true
dropout: 0.5
linear:
hidden_size: 128
num_layers: 2
@@ -1,2 +0,0 @@
# @package _group_
_target_: pyannote.audio.models.embedding.XVectorMFCC
@@ -1,2 +0,0 @@
# @package _group_
_target_: pyannote.audio.models.embedding.XVectorSincNet
@@ -1,7 +0,0 @@
# @package _group_
_target_: torch.optim.Adam
lr: 1e-3
betas: [0.9, 0.999]
eps: 1e-08
weight_decay: 0
amsgrad: False
@@ -1,7 +0,0 @@
# @package _group_
_target_: torch.optim.AdamW
lr: 1e-3
betas: [0.9, 0.999]
eps: 1e-08
weight_decay: 0.01
amsgrad: False
@@ -1,5 +0,0 @@
# @package _group_
_target_: adan_pytorch.Adan
lr: 1e-3
betas: [0.1, 0.1, 0.001]
weight_decay: 0.0
@@ -1,3 +0,0 @@
# @package _group_
_target_: pyannote.audio.utils.preprocessors.LowerTemporalResolution
resolution: 0.1
@@ -1,5 +0,0 @@
# @package _group_
_target_: pyannote.audio.cli.lr_schedulers.CosineAnnealingWarmRestarts
min_lr: 1e-8
max_lr: 1e-3
patience: 1
@@ -1,6 +0,0 @@
# @package _group_
_target_: pyannote.audio.cli.lr_schedulers.CyclicLR
min_lr: 1e-8
max_lr: 1e-3
mode: triangular2
patience: 50
@@ -1,6 +0,0 @@
# @package _group_
_target_: pyannote.audio.cli.lr_schedulers.ReduceLROnPlateau
min_lr: 1e-8
max_lr: 1e-3
factor: 0.5
patience: 50
@@ -1,9 +0,0 @@
# @package _group_
_target_: pyannote.audio.tasks.MultiLabelSegmentation
duration: 3.0
warm_up: 0.0
balance: null
weight: null
batch_size: 32
num_workers: null
pin_memory: False
@@ -1,13 +0,0 @@
# @package _group_
_target_: pyannote.audio.tasks.OverlappedSpeechDetection
duration: 3.0
warm_up: 0.0
balance: null
overlap:
probability: 0.5
snr_min: 0.0
snr_max: 10.0
weight: null
batch_size: 32
num_workers: null
pin_memory: False
@@ -1,8 +0,0 @@
# @package _group_
_target_: pyannote.audio.tasks.SpeakerDiarization
duration: 5.0
max_speakers_per_chunk: 3
max_speakers_per_frame: 2
batch_size: 32
num_workers: 10
pin_memory: False
@@ -1,10 +0,0 @@
# @package _group_
_target_: pyannote.audio.tasks.SupervisedRepresentationLearningWithArcFace
min_duration: 2.0
duration: 5.0
num_classes_per_batch: 512
num_chunks_per_class: 1
margin: 2.0
scale: 12.0
num_workers: null
pin_memory: False
@@ -1,9 +0,0 @@
# @package _group_
_target_: pyannote.audio.tasks.VoiceActivityDetection
duration: 3.0
warm_up: 0.0
balance: null
weight: null
batch_size: 32
num_workers: null
pin_memory: False
@@ -1,35 +0,0 @@
# @package _group_
_target_: pytorch_lightning.Trainer
accelerator: auto
accumulate_grad_batches: 1
benchmark: null # TODO: automatically set to True when using fixed duration chunks
deterministic: False
check_val_every_n_epoch: 1
devices: auto
detect_anomaly: False
enable_checkpointing: True
enable_model_summary: True
enable_progress_bar: True
fast_dev_run: False
gradient_clip_val: null
gradient_clip_algorithm: norm
limit_predict_batches: 1.0
limit_test_batches: 1.0
limit_train_batches: 1.0
limit_val_batches: 1.0
log_every_n_steps: 50
max_epochs: 1000
max_steps: -1
max_time: null
min_epochs: 1
min_steps: null
num_nodes: 1
num_sanity_val_steps: 2
overfit_batches: 0.0
precision: 32
profiler: null
reload_dataloaders_every_n_epochs: 0
use_distributed_sampler: True # TODO: check what this does exactly
strategy: auto
sync_batchnorm: False
val_check_interval: 1.0
@@ -1,3 +0,0 @@
# @package _group_
_target_: pytorch_lightning.Trainer
fast_dev_run: True
+72
View File
@@ -0,0 +1,72 @@
[project]
name = "pyannote-audio"
description = "State-of-the-art speaker diarization toolkit"
readme = "README.md"
authors = [
{ name = "Hervé BREDIN", email = "herve@pyannote.ai" }
]
requires-python = ">=3.10"
dynamic = [
"version",
]
dependencies = [
"asteroid-filterbanks>=0.4.0",
"einops>=0.8.1",
"huggingface-hub>=0.28.1",
"lightning>=2.5.0.post0",
"pyannote-core>=6.0.0rc1",
"pyannote-database>=6.0.0rc1",
"pyannote-metrics>=4.0.0rc1",
"pyannote-pipeline>=4.0.0rc1",
"pytorch-metric-learning>=2.8.1",
"rich>=13.9.4",
"safetensors>=0.5.2",
"torch-audiomentations>=0.12.0",
"torch>=2.6.0",
"torchaudio>=2.6.0",
"torchmetrics>=1.6.1",
"soundfile>=0.13.1",
"matplotlib>=3.10.0",
]
[project.scripts]
pyannote-audio = "pyannote.audio.__main__:app"
[project.optional-dependencies]
test = [
"papermill>=2.6.0",
"pytest>=8.3.4",
"speechbrain>=1.0.2",
]
doc = [
"sphinx-rtd-theme>=3.0.2",
"sphinx>=8.1.3",
]
cli = [
"typer>=0.15.1",
]
separation = [
"asteroid>=0.5.2",
"transformers>=4.48.3",
]
dev = [
"ipython>=8.32.0",
]
[build-system]
requires = ["hatchling", "hatch-vcs"]
build-backend = "hatchling.build"
[tool.hatch.version]
source = "vcs"
[tool.hatch.build.targets.wheel]
packages = ["src/pyannote"]
[dependency-groups]
dev = [
"ipykernel>=6.29.5",
]
-20
View File
@@ -1,20 +0,0 @@
asteroid-filterbanks >=0.4
einops >=0.6.0
huggingface_hub >= 0.13.0
lightning >= 2.0.1
omegaconf >=2.1,<3.0
pyannote.core >= 5.0.0
pyannote.database >= 5.1.3
pyannote.metrics >= 3.2.1
pyannote.pipeline >= 3.0.1
pytorch_metric_learning >= 2.1.0
rich >= 12.0.0
safetensors >= 0.4.5
semver >= 3.0.0
soundfile >= 0.12.1
speechbrain >= 1.0.0
tensorboardX >= 2.6
torch >= 2.0.0
torch_audiomentations >= 0.12.0
torchaudio >= 2.2.0
torchmetrics >= 0.11.0
-119
View File
@@ -1,119 +0,0 @@
# This file is used to configure your project.
# Read more about the various options under:
# http://setuptools.readthedocs.io/en/latest/setuptools.html#configuring-setup-using-setup-cfg-files
[metadata]
name = pyannote-audio
description = Neural speaker diarization
author = Herve Bredin
author-email = herve.bredin@irit.fr
license = mit
long-description = file: README.md
long-description-content-type = text/markdown; charset=UTF-8; variant=GFM
# Change if running only on Windows, Mac or Linux (comma-separated)
platforms = Linux, Mac
# Add here all kinds of additional classifiers as defined under
# https://pypi.python.org/pypi?%3Aaction=list_classifiers
classifiers =
Development Status :: 4 - Beta
Programming Language :: Python
[options]
zip_safe = False
packages = find:
include_package_data = True
# DON'T CHANGE THE FOLLOWING LINE! IT WILL BE UPDATED BY PYSCAFFOLD!
setup_requires = pyscaffold>=3.2a0,<3.3a0
# Add here dependencies of your project (semicolon/line-separated), e.g.
# install_requires = numpy; scipy
# Require a specific Python version, e.g. Python 2.7 or >= 3.4
python_requires = >=3.9
[options.packages.find]
where = .
exclude =
tests
[options.extras_require]
# Add here additional requirements for extra features, to install with:
# `pip install fastaudio[PDF]` like:
# PDF = ReportLab; RXP
# Add here test requirements (semicolon/line-separated)
testing =
pytest>=6.0
pytest-cov>=2.10
jupyter
papermill
dev =
pre_commit>=2.7
recommonmark>=0.6
black>=22.3.0
cli =
hydra-core >=1.1,<1.2
typer >= 0.12.0
separation =
transformers >= 4.39.1
asteroid >=0.7.0
[options.entry_points]
console_scripts =
pyannote-audio-train=pyannote.audio.cli.train:train
pyannote-audio-eval=pyannote.audio.cli.evaluate:evaluate
[test]
# py.test options when running `python setup.py test`
# addopts = --verbose
extras = True
[tool:pytest]
# Options for py.test:
# Specify command line options as you would do when invoking py.test directly.
# e.g. --cov-report html (or xml) for html/xml output or --junitxml junit.xml
# in order to write a coverage file that can be read by Jenkins.
addopts =
--cov pyannote --cov-report term-missing
--verbose
norecursedirs =
dist
build
.tox
testpaths = tests
[aliases]
dists = bdist_wheel
[bdist_wheel]
# Use this option if your package is pure-python
universal = 1
[build_sphinx]
source_dir = doc
build_dir = build/sphinx
[devpi:upload]
# Options for the devpi: PyPI server and packaging tool
# VCS export must be deactivated since we are using setuptools-scm
no-vcs = 1
formats = bdist_wheel
[flake8]
# Some sane defaults for the code style checker flake8
exclude =
.tox
build
dist
.eggs
docs/conf.py
[pyscaffold]
# PyScaffold's parameters when the project was created.
# This will be used when updating. Do not change!
version = 3.2.3
package = pyannote-audio
extensions =
markdown
no_skeleton
pre_commit
dsproject
@@ -20,11 +20,10 @@
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
try:
from .version import __version__, git_version # noqa: F401
except ImportError:
pass
import importlib.metadata
__version__ = importlib.metadata.version("pyannote-audio")
from .core.inference import Inference
from .core.io import Audio
File renamed without changes.
File renamed without changes.
File renamed without changes.
File renamed without changes.
@@ -22,8 +22,8 @@
from typing import List, Mapping, Optional, Text, Union
from pytorch_lightning import Callback, Trainer
from pytorch_lightning.utilities.model_summary import ModelSummary
from lightning.pytorch import Callback, Trainer
from lightning.pytorch.utilities.model_summary import ModelSummary
from pyannote.audio import Model
@@ -80,7 +80,6 @@ class GraduallyUnfreeze(Callback):
self.schedule = schedule
def on_fit_start(self, trainer: Trainer, model: Model):
schedule = self.schedule
task_specific_layers = model.task_dependent
@@ -28,8 +28,8 @@ import torch
import torch.nn as nn
import torch.nn.functional as F
from einops import rearrange
from lightning.pytorch.utilities.memory import is_oom_error
from pyannote.core import Segment, SlidingWindow, SlidingWindowFeature
from pytorch_lightning.utilities.memory import is_oom_error
from pyannote.audio.core.io import AudioFile
from pyannote.audio.core.model import Model, Specifications
@@ -34,7 +34,6 @@ from io import IOBase
from pathlib import Path
from typing import Mapping, Optional, Text, Tuple, Union
import numpy as np
import torch.nn.functional as F
import torchaudio
from pyannote.core import Segment
@@ -30,13 +30,13 @@ from importlib import import_module
from pathlib import Path
from typing import Any, Dict, List, Optional, Text, Tuple, Union
import pytorch_lightning as pl
import lightning.pytorch as pl
import torch
import torch.nn as nn
import torch.optim
from lightning_fabric.utilities.cloud_io import _load as pl_load
from lightning.fabric.utilities.cloud_io import _load as pl_load
from lightning.pytorch.utilities.model_summary.model_summary import ModelSummary
from pyannote.core import SlidingWindow
from pytorch_lightning.utilities.model_summary.model_summary import ModelSummary
from torch.utils.data import DataLoader
from pyannote.audio import __version__
File renamed without changes.
@@ -36,8 +36,8 @@ from pathlib import Path
from tempfile import mkstemp
from typing import Dict, List, Literal, Optional, Sequence, Text, Tuple, Union
import lightning.pytorch as pl
import numpy as np
import pytorch_lightning as pl
import scipy.special
import torch
from pyannote.database import Protocol
@@ -872,8 +872,8 @@ class Task(pl.LightningDataModule):
See also
--------
pytorch_lightning.callbacks.ModelCheckpoint
pytorch_lightning.callbacks.EarlyStopping
lightning.pytorch.callbacks.ModelCheckpoint
lightning.pytorch.callbacks.EarlyStopping
"""
name, metric = next(iter(self.metric.items()))
File renamed without changes.
@@ -25,7 +25,7 @@
import sys
from pathlib import Path
import pytorch_lightning as pl
import lightning.pytorch as pl
import torch
import pyannote.audio.models.embedding.wespeaker as wespeaker
@@ -20,7 +20,6 @@
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
import contextlib
from functools import lru_cache
from typing import Optional, Union
@@ -656,8 +656,8 @@ class SpeechSeparation(SpeakerDiarizationMixin, Pipeline):
# separated sources might be scaled up/down due to SI-SDR loss used when training
# so we peak-normalize them
sources.data = (
sources.data / (np.max(np.abs(sources.data), axis=0, keepdims=True) + 1e-8)
sources.data = sources.data / (
np.max(np.abs(sources.data), axis=0, keepdims=True) + 1e-8
)
# convert to continuous diarization
Loaded 100 of 155 files, more files were not shown because too many files have changed in this diff. Show more