mirrors/kaldifeat
1
0
Fork 0
mirror of https://github.com/csukuangfj/kaldifeat.git synced 2025-08-13 20:12:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into fix_android

Browse Source
This commit is contained in:
pkufool 2022-07-04 15:44:51 +08:00
commit a9f2fe9ecb
59 changed files with 2366 additions and 549 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
.github/workflows/build-doc.yml vendored Normal file
View File

@ -0,0 +1,62 @@
# Copyright 2022 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# refer to https://github.com/actions/starter-workflows/pull/47/files
# You can access it at https://csukuangfj.github.io/kaldifeat
name: Generate doc
on:
push:
branches:
- master
- doc
jobs:
build-doc:
runs-on: ${{ matrix.os }}
strategy:
fail-fast: false
matrix:
os: [ubuntu-latest]
python-version: [3.8]
steps:
# refer to https://github.com/actions/checkout
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Setup Python ${{ matrix.python-version }}
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Display Python version
run: python -c "import sys; print(sys.version)"
- name: Build doc
shell: bash
run: |
cd doc
python3 -m pip install -r ./requirements.txt
make html
touch build/html/.nojekyll
- name: Deploy
uses: peaceiris/actions-gh-pages@v3
with:
github_token: ${{ secrets.GITHUB_TOKEN }}
publish_dir: ./doc/build/html
publish_branch: gh-pages

108
.github/workflows/build_conda_ubuntu_cpu.yml vendored Normal file
View File

@ -0,0 +1,108 @@
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# refer to https://github.com/actions/starter-workflows/pull/47/files
name: build_conda_ubuntu_cpu
on:
push:
tags:
- '*'
env:
KALDIFEAT_BUILD_TYPE: Release
jobs:
generate_build_matrix:
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py
MATRIX=$(python scripts/github_actions/generate_build_matrix.py)
echo "::set-output name=matrix::${MATRIX}"
build_conda_ubuntu_cpu:
needs: generate_build_matrix
runs-on: ubuntu-18.04
strategy:
fail-fast: false
matrix:
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
# refer to https://github.com/actions/checkout
- uses: actions/checkout@v2
with:
fetch-depth: 0
- uses: conda-incubator/setup-miniconda@v2
with:
auto-update-conda: true
python-version: ${{ matrix.python-version }}
activate-environment: kaldifeat
- name: Display Python version
shell: bash -l {0}
run: |
python3 -c "import sys; print(sys.version)"
which python3
- name: Install conda dependencies
shell: bash -l {0}
run: |
conda install -y -q anaconda-client
conda install -y -q conda-build
conda install -y -q -c pytorch pytorch=${{ matrix.torch }} cpuonly
- name: Display conda info
shell: bash -l {0}
run: |
which conda
conda env list
conda info
nproc
- name: Build kaldifeat
shell: bash -l {0}
env:
KALDIFEAT_PYTHON_VERSION: ${{ matrix.python-version}}
KALDIFEAT_TORCH_VERSION: ${{ matrix.torch }}
KALDIFEAT_CONDA_TOKEN: ${{ secrets.KALDIFEAT_CONDA_TOKEN}}
KALDIFEAT_IS_GITHUB_ACTIONS: 1
KALDIFEAT_IS_FOR_CONDA: 1
run: |
export KALDIFEAT_BUILD_TYPE=$KALDIFEAT_BUILD_TYPE
./scripts/build_conda_cpu.sh
- name: Display generated files
run: |
ls -lh /usr/share/miniconda/envs/kaldifeat/conda-bld/linux-64
- name: Upload generated files
uses: actions/upload-artifact@v2
with:
name: cpu-torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-${{ matrix.os }}
path: /usr/share/miniconda/envs/kaldifeat/conda-bld/linux-64/*.tar.bz2

95
.github/workflows/build_conda.yml → .github/workflows/build_conda_ubuntu_cuda.yml vendored
View File

@ -16,7 +16,7 @@
# refer to https://github.com/actions/starter-workflows/pull/47/files
name: build_conda_cuda
name: build_conda_ubuntu_cuda
on:
push:
@ -27,70 +27,30 @@ env:
KALDIFEAT_BUILD_TYPE: Release
jobs:
build_conda_cuda:
runs-on: ${{ matrix.os }}
generate_build_matrix:
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py --enable-cuda
MATRIX=$(python scripts/github_actions/generate_build_matrix.py --enable-cuda)
echo "::set-output name=matrix::${MATRIX}"
build_conda_ubuntu_cuda:
needs: generate_build_matrix
runs-on: ubuntu-18.04
strategy:
fail-fast: false
matrix:
os: [ubuntu-18.04]
# anaconda does not support 3.9 as of 2021.05.08
python-version: [3.6, 3.7, 3.8, 3.9]
# python-version: [3.6, 3.7, 3.8]
cuda: ["10.1", "10.2", "11.0", "11.1"]
# from https://download.pytorch.org/whl/torch_stable.html
#
# PyTorch 1.9.0 supports: 10.2 (default), 11.1
# PyTorch 1.8.1 supports: cuda 10.1, 10.2 (default), 11.1
# PyTorch 1.8.0 supports: cuda 10.1, 10.2 (default), 11.1
# PyTorch 1.7.x supports: cuda 10.1, 10.2 (default), 11.0, 9.2 (not included in this setup)
# PyTorch 1.6.0 supports: cuda 10.1, 10.2 (default), 9.2 (not included in this setup)
# PyTorch 1.5.x supports: cuda 10.1, 10.2 (default), 9.2 (not included in this setup)
#
# PyTorch 1.8.x and 1.7.1 support 3.6, 3.7, 3.8, 3.9
# PyTorch 1.7.0, 1.6.0, and 1.5.x support 3.6, 3.7, 3.8
#
# Other PyTorch versions are not tested
#
# torch: ["1.5.0", "1.5.1", "1.6.0", "1.7.0", "1.7.1", "1.8.0", "1.8.1"]
# 1.5.x is removed because there are compilation errors.
# See
# https://github.com/csukuangfj/k2/runs/2533830771?check_suite_focus=true
# and
# https://github.com/NVIDIA/apex/issues/805
torch: ["1.6.0", "1.7.0", "1.7.1", "1.8.0", "1.8.1", "1.9.0"]
exclude:
# - cuda: "11.0" # exclude 11.0 for [1.5.0, 1.5.1, 1.6.0, 1.8.0, 1.8.1, 1.9.0]
# torch: "1.5.0"
# - cuda: "11.0"
# torch: "1.5.1"
- cuda: "11.0"
torch: "1.6.0"
- cuda: "11.0"
torch: "1.8.0"
- cuda: "11.0"
torch: "1.8.1"
- cuda: "11.0"
torch: "1.9.0"
# - cuda: "11.1" # exclude 11.1 for [1.5.0, 1.5.1, 1.6.0, 1.7.0, 1.7.1]
# torch: "1.5.0"
# - cuda: "11.1"
# torch: "1.5.1"
- cuda: "11.1"
torch: "1.6.0"
- cuda: "11.1"
torch: "1.7.0"
- cuda: "11.1"
torch: "1.7.1"
- cuda: "10.1" # exclude 10.1 for [1.9.0]
torch: "1.9.0"
- python-version: 3.9 # exclude Python 3.9 for [1.5.0, 1.5.1, 1.6.0, 1.7.0]
torch: "1.5.0"
- python-version: 3.9
torch: "1.5.1"
- python-version: 3.9
torch: "1.6.0"
- python-version: 3.9
torch: "1.7.0"
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
# refer to https://github.com/actions/checkout
@ -131,7 +91,6 @@ jobs:
run: |
conda install -y -q anaconda-client
conda install -y -q conda-build
conda install -y -q bs4 requests tqdm
conda install -y -q -c pytorch -c conda-forge pytorch=${{ matrix.torch }} cudatoolkit=${{ matrix.cuda }}
- name: Display conda info
@ -161,3 +120,13 @@ jobs:
run: |
export KALDIFEAT_BUILD_TYPE=$KALDIFEAT_BUILD_TYPE
./scripts/build_conda.sh
- name: Display generated files
run: |
ls -lh /usr/share/miniconda/envs/kaldifeat/conda-bld/linux-64
- name: Upload generated files
uses: actions/upload-artifact@v2
with:
name: cuda-${{ matrix.cuda }}-torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-${{ matrix.os }}
path: /usr/share/miniconda/envs/kaldifeat/conda-bld/linux-64/*.tar.bz2

116
.github/workflows/nightly-tests-macos-cpu.yml vendored Normal file
View File

@ -0,0 +1,116 @@
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
name: Nightly tests macos cpu
on:
schedule:
# minute (0-59)
# hour (0-23)
# day of the month (1-31)
# month (1-12)
# day of the week (0-6)
# nightly build at 23:50 UTC time every day
- cron: "50 23 * * *"
jobs:
enable_nightly_build:
runs-on: ubuntu-latest
outputs:
enabled: ${{ steps.set-enabled.outputs.enabled }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Set enabled
id: set-enabled
run: |
enabled=$(python scripts/github_actions/run-nightly-build.py)
echo "enabled: $enabled"
echo "::set-output name=enabled::${enabled}"
generate_build_matrix:
needs: enable_nightly_build
if: needs.enable_nightly_build.outputs.enabled == 'true'
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py
MATRIX=$(python scripts/github_actions/generate_build_matrix.py)
echo "::set-output name=matrix::${MATRIX}"
run_nightly_tests_macos_cpu:
needs: generate_build_matrix
runs-on: macos-10.15
strategy:
fail-fast: false
matrix:
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Setup Python
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Display Python version
run: python -c "import sys; print(sys.version)"
- name: Display clang version
run: |
clang --version
- name: Install PyTorch ${{ matrix.torch }}
shell: bash
run: |
python3 -m pip install -qq --upgrade pip
python3 -m pip install -qq wheel twine numpy typing_extensions
python3 -m pip install -qq torch==${{ matrix.torch }} numpy
python3 -m pip install -qq dataclasses soundfile
python3 -c "import torch; print('torch version:', torch.__version__)"
- name: Install from PyPI
shell: bash
run: |
export KALDIFEAT_MAKE_ARGS="-j 3"
pip install --verbose kaldifeat
- name: Run tests
shell: bash
run: |
lib_path=$(python -c "from distutils.sysconfig import get_python_lib; print(get_python_lib())")
echo "lib_path: $lib_path"
ls -lh $lib_path
# export DYLD_LIBRARY_PATH=$lib_path:$DYLD_LIBRARY_PATH
otool -L $lib_path/_kaldifeat.*.so
otool -l $lib_path/_kaldifeat.*.so
cd kaldifeat/python/tests
make test

100
.github/workflows/nightly-tests-ubuntu-conda-cpu.yml vendored Normal file
View File

@ -0,0 +1,100 @@
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
name: Nightly tests ubuntu conda cpu
on:
schedule:
# minute (0-59)
# hour (0-23)
# day of the month (1-31)
# month (1-12)
# day of the week (0-6)
# nightly build at 23:50 UTC time every day
- cron: "50 23 * * *"
jobs:
enable_nightly_build:
runs-on: ubuntu-latest
outputs:
enabled: ${{ steps.set-enabled.outputs.enabled }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Set enabled
id: set-enabled
run: |
enabled=$(python scripts/github_actions/run-nightly-build.py)
echo "enabled: $enabled"
echo "::set-output name=enabled::${enabled}"
generate_build_matrix:
needs: enable_nightly_build
if: needs.enable_nightly_build.outputs.enabled == 'true'
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py --test-only-latest-torch
MATRIX=$(python scripts/github_actions/generate_build_matrix.py --test-only-latest-torch)
echo "::set-output name=matrix::${MATRIX}"
run_nightly_tests_ubuntu_conda_cpu:
needs: generate_build_matrix
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- uses: conda-incubator/setup-miniconda@v2
with:
auto-update-conda: true
python-version: ${{ matrix.python-version }}
activate-environment: kaldifeat
- name: Display Python version
shell: bash -l {0}
run: |
python3 -c "import sys; print(sys.version)"
which python3
- name: Install kaldifeat and run tests
shell: bash -l {0}
run: |
sudo apt-get update
sudo apt-get install -y libsndfile1-dev libsndfile1 ffmpeg
conda install -y -c kaldifeat -c pytorch cpuonly pytorch=${{ matrix.torch }} kaldifeat python=${{ matrix.python-version }}
conda install -y -c conda-forge pysoundfile
cd kaldifeat/python/tests
make test

96
.github/workflows/nightly-tests-ubuntu-conda-cuda.yml vendored Normal file
View File

@ -0,0 +1,96 @@
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
name: Nightly tests ubuntu conda cuda
on:
schedule:
# minute (0-59)
# hour (0-23)
# day of the month (1-31)
# month (1-12)
# day of the week (0-6)
# nightly build at 23:50 UTC time every day
- cron: "50 23 * * *"
jobs:
enable_nightly_build:
runs-on: ubuntu-latest
outputs:
enabled: ${{ steps.set-enabled.outputs.enabled }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Set enabled
id: set-enabled
run: |
enabled=$(python scripts/github_actions/run-nightly-build.py)
echo "enabled: $enabled"
echo "::set-output name=enabled::${enabled}"
generate_build_matrix:
needs: enable_nightly_build
if: needs.enable_nightly_build.outputs.enabled == 'true'
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py --enable-cuda --test-only-latest-torch
MATRIX=$(python scripts/github_actions/generate_build_matrix.py --enable-cuda --test-only-latest-torch)
echo "::set-output name=matrix::${MATRIX}"
run_nightly_tests_ubuntu_conda_cuda:
needs: generate_build_matrix
runs-on: ubuntu-latest
strategy:
fail-fast: false
matrix:
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- uses: conda-incubator/setup-miniconda@v2
with:
auto-update-conda: true
python-version: ${{ matrix.python-version }}
activate-environment: kaldifeat
- name: Display Python version
shell: bash -l {0}
run: |
python3 -c "import sys; print(sys.version)"
which python3
- name: Install kaldifeat and run tests
shell: bash -l {0}
run: |
conda install -y -c kaldifeat -c pytorch -c conda-forge pytorch=${{ matrix.torch }} cudatoolkit=${{ matrix.cuda }} kaldifeat python=${{ matrix.python-version }}
conda install -y -c conda-forge pysoundfile
cd kaldifeat/python/tests
make test

118
.github/workflows/nightly-tests-ubuntu-pip-cpu.yml vendored Normal file
View File

@ -0,0 +1,118 @@
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
name: Nightly tests ubuntu pip cpu
on:
schedule:
# minute (0-59)
# hour (0-23)
# day of the month (1-31)
# month (1-12)
# day of the week (0-6)
# nightly build at 23:50 UTC time every day
- cron: "50 23 * * *"
jobs:
enable_nightly_build:
runs-on: ubuntu-latest
outputs:
enabled: ${{ steps.set-enabled.outputs.enabled }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Set enabled
id: set-enabled
run: |
enabled=$(python scripts/github_actions/run-nightly-build.py)
echo "enabled: $enabled"
echo "::set-output name=enabled::${enabled}"
generate_build_matrix:
needs: enable_nightly_build
if: needs.enable_nightly_build.outputs.enabled == 'true'
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py
MATRIX=$(python scripts/github_actions/generate_build_matrix.py)
echo "::set-output name=matrix::${MATRIX}"
run_nightly_tests_ubuntu_pip_cpu:
needs: generate_build_matrix
runs-on: ubuntu-18.04
strategy:
fail-fast: false
matrix:
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Setup Python
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Display Python version
run: python -c "import sys; print(sys.version)"
- name: Install GCC 7
run: |
sudo apt-get install -y gcc-7 g++-7
echo "CC=/usr/bin/gcc-7" >> $GITHUB_ENV
echo "CXX=/usr/bin/g++-7" >> $GITHUB_ENV
- name: Display GCC version
run: |
gcc --version
- name: Install PyTorch ${{ matrix.torch }}
shell: bash
run: |
sudo apt-get update
sudo apt-get install -y libsndfile1-dev libsndfile1 ffmpeg
python3 -m pip install -qq --upgrade pip
python3 -m pip install -qq wheel twine typing_extensions numpy soundfile
python3 -m pip install -qq torch==${{ matrix.torch }}+cpu -f https://download.pytorch.org/whl/torch_stable.html
python3 -m pip install -qq dataclasses
python3 -c "import torch; print('torch version:', torch.__version__)"
- name: Install from PyPI
shell: bash
run: |
export KALDIFEAT_MAKE_ARGS="-j 3"
pip install --verbose kaldifeat
- name: Run tests
shell: bash
run: |
cd kaldifeat/python/tests
make test

105
.github/workflows/publish_to_pypi.yml-bak → .github/workflows/nightly-tests-ubuntu-pip-cuda.yml vendored
View File

@ -14,23 +14,61 @@
# See the License for the specific language governing permissions and
# limitations under the License.
#
name: Publish to PyPI
name: Nightly tests ubuntu pip cuda
on:
push:
tags:
- '*'
schedule:
# minute (0-59)
# hour (0-23)
# day of the month (1-31)
# month (1-12)
# day of the week (0-6)
# nightly build at 23:50 UTC time every day
- cron: "50 23 * * *"
jobs:
pypi:
runs-on: ${{ matrix.os }}
enable_nightly_build:
runs-on: ubuntu-latest
outputs:
enabled: ${{ steps.set-enabled.outputs.enabled }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Set enabled
id: set-enabled
run: |
enabled=$(python scripts/github_actions/run-nightly-build.py)
echo "enabled: $enabled"
echo "::set-output name=enabled::${enabled}"
generate_build_matrix:
needs: enable_nightly_build
if: needs.enable_nightly_build.outputs.enabled == 'true'
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py --enable-cuda
MATRIX=$(python scripts/github_actions/generate_build_matrix.py --enable-cuda)
echo "::set-output name=matrix::${MATRIX}"
run_nightly_tests_ubuntu_pip_cuda:
needs: generate_build_matrix
runs-on: ubuntu-18.04
strategy:
fail-fast: false
matrix:
os: [ubuntu-18.04, macos-10.15]
cuda: ["10.1"]
torch: ["1.7.1"]
python-version: [3.6, 3.7, 3.8]
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2
with:
@ -42,15 +80,12 @@ jobs:
python-version: ${{ matrix.python-version }}
- name: Install GCC 7
if: startsWith(matrix.os, 'ubuntu')
run: |
sudo apt-get install -y gcc-7 g++-7
echo "CC=/usr/bin/gcc-7" >> $GITHUB_ENV
echo "CXX=/usr/bin/g++-7" >> $GITHUB_ENV
- name: Install CUDA Toolkit ${{ matrix.cuda }}
if: startsWith(matrix.os, 'ubuntu')
shell: bash
env:
cuda: ${{ matrix.cuda }}
@ -61,61 +96,39 @@ jobs:
echo "LD_LIBRARY_PATH=${CUDA_HOME}/lib:${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}" >> $GITHUB_ENV
- name: Display NVCC version
if: startsWith(matrix.os, 'ubuntu')
run: |
which nvcc
nvcc --version
- name: Install PyTorch ${{ matrix.torch }}
if: startsWith(matrix.os, 'ubuntu')
env:
cuda: ${{ matrix.cuda }}
torch: ${{ matrix.torch }}
shell: bash
run: |
sudo apt-get update
sudo apt-get install -y libsndfile1-dev libsndfile1 ffmpeg
python3 -m pip install --upgrade pip
python3 -m pip install wheel twine typing_extensions
python3 -m pip install bs4 requests tqdm
python3 -m pip install wheel twine typing_extensions soundfile
python3 -m pip install bs4 requests tqdm numpy
./scripts/github_actions/install_torch.sh
python3 -c "import torch; print('torch version:', torch.__version__)"
- name: Install PyTorch ${{ matrix.torch }}
if: startsWith(matrix.os, 'macos')
shell: bash
run: |
python3 -m pip install -qq --upgrade pip
python3 -m pip install -qq wheel twine typing_extensions
python3 -m pip install -qq torch==${{ matrix.torch }}
python3 -c "import torch; print('torch version:', torch.__version__)"
- name: Download cudnn 8.0
if: startsWith(matrix.os, 'ubuntu')
env:
cuda: ${{ matrix.cuda }}
run: |
./scripts/github_actions/install_cudnn.sh
- name: Build pip packages
- name: Install from PyPI
shell: bash
env:
KALDIFEAT_IS_FOR_PYPI: 1
run: |
tag=$(python3 -c "import sys; print(''.join(sys.version[:3].split('.')))")
export KALDIFEAT_MAKE_ARGS="-j2"
python3 setup.py bdist_wheel --python-tag=py${tag}
ls -lh dist/
export KALDIFEAT_MAKE_ARGS="-j 3"
pip install --verbose kaldifeat
- name: Publish wheels to PyPI
env:
TWINE_USERNAME: ${{ secrets.PYPI_USERNAME }}
TWINE_PASSWORD: ${{ secrets.PYPI_PASSWORD }}
- name: Run tests
shell: bash
run: |
twine upload dist/kaldifeat-*.whl
- name: Upload Wheel
uses: actions/upload-artifact@v2
with:
name: torch-${{ matrix.torch }}-python-${{ matrix.python-version }}-${{ matrix.os }}
path: dist/*.whl
cd kaldifeat/python/tests
make test

119
.github/workflows/nightly-tests-windows-pip-cpu.yml vendored Normal file
View File

@ -0,0 +1,119 @@
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
name: Nightly tests windows pip cpu
on:
schedule:
# minute (0-59)
# hour (0-23)
# day of the month (1-31)
# month (1-12)
# day of the week (0-6)
# nightly build at 23:50 UTC time every day
- cron: "50 23 * * *"
jobs:
enable_nightly_build:
runs-on: ubuntu-latest
outputs:
enabled: ${{ steps.set-enabled.outputs.enabled }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Set enabled
id: set-enabled
run: |
enabled=$(python scripts/github_actions/run-nightly-build.py)
echo "enabled: $enabled"
echo "::set-output name=enabled::${enabled}"
generate_build_matrix:
needs: enable_nightly_build
if: needs.enable_nightly_build.outputs.enabled == 'true'
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py
MATRIX=$(python scripts/github_actions/generate_build_matrix.py)
echo "::set-output name=matrix::${MATRIX}"
run_nightly_tests_windows_pip_cpu:
needs: generate_build_matrix
runs-on: windows-2019
strategy:
fail-fast: false
matrix:
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
# see https://github.com/microsoft/setup-msbuild
- name: Add msbuild to PATH
uses: microsoft/setup-msbuild@v1.0.2
- name: Setup Python
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Display Python version
run: python -c "import sys; print(sys.version)"
- name: Install PyTorch ${{ matrix.torch }}
shell: bash
run: |
python3 -m pip install -qq --upgrade pip
python3 -m pip install -qq wheel twine numpy typing_extensions
python3 -m pip install -qq dataclasses soundfile numpy
python3 -m pip install -qq torch==${{ matrix.torch }}+cpu -f https://download.pytorch.org/whl/torch_stable.html
python3 -c "import torch; print('torch version:', torch.__version__)"
- name: Install from PyPI
shell: bash
run: |
export KALDIFEAT_MAKE_ARGS="-j 3"
pip install --verbose kaldifeat
- name: Run tests
shell: bash
run: |
cd kaldifeat/python/tests
python3 ./test_fbank.py
python3 ./test_fbank_options.py
python3 ./test_frame_extraction_options.py
python3 ./test_mel_bank_options.py
python3 ./test_mfcc.py
python3 ./test_mfcc_options.py
python3 ./test_plp.py
python3 ./test_plp_options.py
python3 ./test_spectrogram.py
python3 ./test_spectrogram_options.py

160
.github/workflows/nightly-tests-windows-pip-cuda.yml vendored Normal file
View File

@ -0,0 +1,160 @@
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
name: Nightly tests windows pip cuda
on:
schedule:
# minute (0-59)
# hour (0-23)
# day of the month (1-31)
# month (1-12)
# day of the week (0-6)
# nightly build at 23:50 UTC time every day
- cron: "50 23 * * *"
jobs:
enable_nightly_build:
runs-on: ubuntu-latest
outputs:
enabled: ${{ steps.set-enabled.outputs.enabled }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Set enabled
id: set-enabled
run: |
enabled=$(python scripts/github_actions/run-nightly-build.py)
echo "enabled: $enabled"
echo "::set-output name=enabled::${enabled}"
generate_build_matrix:
needs: enable_nightly_build
if: needs.enable_nightly_build.outputs.enabled == 'true'
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py --enable-cuda --for-windows
MATRIX=$(python scripts/github_actions/generate_build_matrix.py --enable-cuda --for-windows)
echo "::set-output name=matrix::${MATRIX}"
run_nightly_tests_windows_pip_cuda:
needs: generate_build_matrix
runs-on: windows-2019
strategy:
fail-fast: false
matrix:
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
# see https://github.com/microsoft/setup-msbuild
- name: Add msbuild to PATH
uses: microsoft/setup-msbuild@v1.0.2
- name: Setup Python
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Display Python version
run: python -c "import sys; print(sys.version)"
# See https://github.com/Jimver/cuda-toolkit/blob/master/src/links/windows-links.ts
# for available CUDA versions
- uses: Jimver/cuda-toolkit@v0.2.7
id: cuda-toolkit
with:
cuda: ${{ matrix.cuda }}
- name: Display CUDA version
run: |
echo "Installed cuda version is: ${{ steps.cuda-toolkit.outputs.cuda }}"
echo "Cuda install location: ${{steps.cuda-toolkit.outputs.CUDA_PATH}}"
nvcc --version
- name: Remove CUDA installation package
shell: bash
run: |
rm "C:/hostedtoolcache/windows/cuda_installer-windows/${{ matrix.cuda }}/x64/cuda_installer_${{ matrix.cuda }}.exe"
- name: Download cuDNN
shell: bash
run: |
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/csukuangfj/cudnn-for-windows
cd cudnn-for-windows
git lfs pull --include="cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive.zip"
unzip cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive.zip
rm cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive.zip
ls -lh *
ls -lh */*
echo "PWD: $PWD"
- name: Install PyTorch ${{ matrix.torch }}
shell: bash
run: |
version=${{ matrix.cuda }}
major=${version:0:2}
minor=${version:3:1}
v=${major}${minor}
if [ ${v} -eq 102 ]; then v=""; else v="+cu${v}"; fi
python3 -m pip install -qq --upgrade pip
python3 -m pip install -qq wheel twine numpy typing_extensions
python3 -m pip install -qq dataclasses soundfile numpy
python3 -m pip install -qq torch==${{ matrix.torch }}${v} -f https://download.pytorch.org/whl/torch_stable.html numpy
python3 -c "import torch; print('torch version:', torch.__version__)"
python3 -m torch.utils.collect_env
- name: Install from PyPI
shell: bash
run: |
echo "PWD: $PWD"
export KALDIFEAT_CMAKE_ARGS="-DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCUDNN_INCLUDE_PATH=d:/a/kaldifeat/kaldifeat/cudnn-for-windows/cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive/include -DCUDNN_LIBRARY_PATH=d:/a/kaldifeat/kaldifeat/cudnn-for-windows/cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive/lib/cudnn.lib"
export KALDIFEAT_MAKE_ARGS="-j 3"
pip install --verbose kaldifeat
- name: Run tests
shell: bash
run: |
cd kaldifeat/python/tests
python3 ./test_fbank.py
python3 ./test_fbank_options.py
python3 ./test_frame_extraction_options.py
python3 ./test_mel_bank_options.py
python3 ./test_mfcc.py
python3 ./test_mfcc_options.py
python3 ./test_plp.py
python3 ./test_plp_options.py
python3 ./test_spectrogram.py
python3 ./test_spectrogram_options.py

85
.github/workflows/run-tests-macos-cpu.yml vendored Normal file
View File

@ -0,0 +1,85 @@
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
name: Run tests macos cpu
on:
push:
branches:
- master
pull_request:
branches:
- master
jobs:
generate_build_matrix:
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py --test-only-latest-torch
MATRIX=$(python scripts/github_actions/generate_build_matrix.py --test-only-latest-torch)
echo "::set-output name=matrix::${MATRIX}"
run_tests_macos_cpu:
needs: generate_build_matrix
runs-on: macos-10.15
strategy:
fail-fast: false
matrix:
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Setup Python
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Install PyTorch ${{ matrix.torch }}
shell: bash
run: |
python3 -m pip install -qq --upgrade pip
python3 -m pip install -qq wheel twine typing_extensions soundfile numpy
python3 -m pip install -qq torch==${{ matrix.torch }} -f https://download.pytorch.org/whl/torch_stable.html
python3 -c "import torch; print('torch version:', torch.__version__)"
- name: Build
shell: bash
run: |
mkdir build_release
cd build_release
cmake ..
make VERBOSE=1 -j3
- name: Run tests
shell: bash
run: |
cd build_release
ctest --output-on-failure

94
.github/workflows/run-tests-ubuntu-cpu.yml vendored Normal file
View File

@ -0,0 +1,94 @@
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
name: Run tests ubuntu cpu
on:
push:
branches:
- master
pull_request:
branches:
- master
jobs:
generate_build_matrix:
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py --test-only-latest-torch
MATRIX=$(python scripts/github_actions/generate_build_matrix.py --test-only-latest-torch)
echo "::set-output name=matrix::${MATRIX}"
run_tests_ubuntu_cpu:
needs: generate_build_matrix
runs-on: ubuntu-18.04
strategy:
fail-fast: false
matrix:
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Setup Python
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Install GCC 7
run: |
sudo apt-get install -y gcc-7 g++-7
echo "CC=/usr/bin/gcc-7" >> $GITHUB_ENV
echo "CXX=/usr/bin/g++-7" >> $GITHUB_ENV
- name: Install PyTorch ${{ matrix.torch }}
shell: bash
run: |
sudo apt-get update
sudo apt-get install -y libsndfile1-dev libsndfile1 ffmpeg
python3 -m pip install --upgrade pip
python3 -m pip install wheel twine typing_extensions soundfile
python3 -m pip install bs4 requests tqdm numpy
python3 -m pip install -qq torch==${{ matrix.torch }}+cpu -f https://download.pytorch.org/whl/torch_stable.html
python3 -c "import torch; print('torch version:', torch.__version__)"
- name: Build
shell: bash
run: |
mkdir build_release
cd build_release
cmake ..
make VERBOSE=1 -j3
- name: Run tests
shell: bash
run: |
cd build_release
ctest --output-on-failure

55
.github/workflows/run-tests.yml → .github/workflows/run-tests-ubuntu-cuda.yml vendored
View File

@ -14,26 +14,43 @@
# See the License for the specific language governing permissions and
# limitations under the License.
#
name: Run tests
name: Run tests ubuntu cuda
on:
push:
branches:
- master
pull_request:
branches:
- master
jobs:
run_tests:
runs-on: ${{ matrix.os }}
generate_build_matrix:
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py --enable-cuda --test-only-latest-torch
MATRIX=$(python scripts/github_actions/generate_build_matrix.py --enable-cuda --test-only-latest-torch)
echo "::set-output name=matrix::${MATRIX}"
run_tests_ubuntu_cuda:
needs: generate_build_matrix
runs-on: ubuntu-18.04
strategy:
fail-fast: false
matrix:
os: [ubuntu-18.04, macos-10.15]
cuda: ["10.1"]
torch: ["1.7.1"]
python-version: [3.6, 3.7, 3.8]
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2
with:
@ -53,7 +70,6 @@ jobs:
- name: Install CUDA Toolkit ${{ matrix.cuda }}
if: startsWith(matrix.os, 'ubuntu')
shell: bash
env:
cuda: ${{ matrix.cuda }}
@ -64,39 +80,26 @@ jobs:
echo "LD_LIBRARY_PATH=${CUDA_HOME}/lib:${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}" >> $GITHUB_ENV
- name: Display NVCC version
if: startsWith(matrix.os, 'ubuntu')
run: |
which nvcc
nvcc --version
- name: Install PyTorch ${{ matrix.torch }}
if: startsWith(matrix.os, 'ubuntu')
env:
cuda: ${{ matrix.cuda }}
torch: ${{ matrix.torch }}
shell: bash
run: |
sudo apt update
sudo apt install libsndfile1-dev libsndfile1 ffmpeg
sudo apt-get update
sudo apt-get install -y libsndfile1-dev libsndfile1 ffmpeg
python3 -m pip install --upgrade pip
python3 -m pip install wheel twine typing_extensions soundfile
python3 -m pip install bs4 requests tqdm
python3 -m pip install bs4 requests tqdm numpy
./scripts/github_actions/install_torch.sh
python3 -c "import torch; print('torch version:', torch.__version__)"
- name: Install PyTorch ${{ matrix.torch }}
if: startsWith(matrix.os, 'macos')
shell: bash
run: |
python3 -m pip install -qq --upgrade pip
python3 -m pip install -qq wheel twine typing_extensions soundfile
python3 -m pip install -qq torch==${{ matrix.torch }}
python3 -c "import torch; print('torch version:', torch.__version__)"
- name: Download cudnn 8.0
if: startsWith(matrix.os, 'ubuntu')
env:
cuda: ${{ matrix.cuda }}
run: |
@ -108,10 +111,10 @@ jobs:
mkdir build_release
cd build_release
cmake ..
make VERBOSE=1
make VERBOSE=1 -j3
- name: Run tests
shell: bash
run: |
cd build_release
ctest -R py --output-on-failure
ctest --output-on-failure

45
.github/workflows/build_windows.yml → .github/workflows/run-tests-windows-cpu.yml vendored
View File

@ -1,23 +1,56 @@
name: build
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
name: Run tests windows cpu
on:
push:
branches:
- master
pull_request:
branches:
- master
jobs:
build-windows:
generate_build_matrix:
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py --test-only-latest-torch
MATRIX=$(python scripts/github_actions/generate_build_matrix.py --test-only-latest-torch)
echo "::set-output name=matrix::${MATRIX}"
run_tests_windows_cpu:
# see https://github.com/actions/virtual-environments/blob/win19/20210525.0/images/win/Windows2019-Readme.md
runs-on: ${{ matrix.os }}
needs: generate_build_matrix
runs-on: windows-2019
strategy:
fail-fast: false
matrix:
os: [windows-2019]
torch: ["1.6.0", "1.7.0", "1.7.1", "1.8.0", "1.8.1", "1.9.0", "1.10.0", "1.11.0"]
python-version: [3.6, 3.7, 3.8, 3.9]
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2

173
.github/workflows/run-tests-windows-cuda.yml vendored Normal file
View File

@ -0,0 +1,173 @@
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
# See ../../LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
name: Run tests windows cuda
on:
push:
branches:
- master
pull_request:
branches:
- master
jobs:
generate_build_matrix:
# see https://github.com/pytorch/pytorch/pull/50633
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.set-matrix.outputs.matrix }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
- name: Generating build matrix
id: set-matrix
run: |
# outputting for debugging purposes
python scripts/github_actions/generate_build_matrix.py --enable-cuda --for-windows --test-only-latest-torch
MATRIX=$(python scripts/github_actions/generate_build_matrix.py --enable-cuda --for-windows --test-only-latest-torch)
echo "::set-output name=matrix::${MATRIX}"
run_tests_windows_cuda:
needs: generate_build_matrix
runs-on: windows-2019
strategy:
fail-fast: false
matrix:
${{ fromJson(needs.generate_build_matrix.outputs.matrix) }}
steps:
- uses: actions/checkout@v2
with:
fetch-depth: 0
# see https://github.com/microsoft/setup-msbuild
- name: Add msbuild to PATH
uses: microsoft/setup-msbuild@v1.0.2
- name: Setup Python ${{ matrix.python-version }}
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Display Python version
run: python -c "import sys; print(sys.version)"
# See https://github.com/Jimver/cuda-toolkit/blob/master/src/links/windows-links.ts
# for available CUDA versions
- uses: Jimver/cuda-toolkit@v0.2.7
id: cuda-toolkit
with:
cuda: ${{ matrix.cuda }}
- name: Display CUDA version
shell: bash
run: |
echo "Installed cuda version is: ${{ steps.cuda-toolkit.outputs.cuda }}"
echo "Cuda install location: ${{steps.cuda-toolkit.outputs.CUDA_PATH}}"
nvcc --version
- name: Remove CUDA installation package
shell: bash
run: |
rm "C:/hostedtoolcache/windows/cuda_installer-windows/${{ matrix.cuda }}/x64/cuda_installer_${{ matrix.cuda }}.exe"
- name: Download cuDNN
shell: bash
run: |
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/csukuangfj/cudnn-for-windows
cd cudnn-for-windows
git lfs pull --include="cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive.zip"
unzip cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive.zip
rm cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive.zip
ls -lh *
ls -lh */*
echo "PWD: $PWD"
- name: Install PyTorch ${{ matrix.torch }}
shell: bash
run: |
version=${{ matrix.cuda }}
major=${version:0:2}
minor=${version:3:1}
v=${major}${minor}
if [ ${v} -eq 102 ]; then v=""; else v="+cu${v}"; fi
python3 -m pip install -qq --upgrade pip
python3 -m pip install -qq wheel twine numpy typing_extensions
python3 -m pip install -qq dataclasses soundfile numpy
python3 -m pip install -qq torch==${{ matrix.torch }}${v} -f https://download.pytorch.org/whl/torch_stable.html numpy
python3 -c "import torch; print('torch version:', torch.__version__)"
python3 -m torch.utils.collect_env
- name: Display CMake version
run: |
cmake --version
cmake --help
- name: Configure CMake
shell: bash
run: |
echo "PWD: $PWD"
ls -lh
mkdir build_release
cd build_release
cmake -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCUDNN_INCLUDE_PATH=d:/a/kaldifeat/kaldifeat/cudnn-for-windows/cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive/include -DCUDNN_LIBRARY_PATH=d:/a/kaldifeat/kaldifeat/cudnn-for-windows/cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive/lib/cudnn.lib ..
ls -lh
- name: Build kaldifeat
shell: bash
run: |
cd build_release
cmake --build . --target _kaldifeat --config Release
- name: Display generated files
shell: bash
run: |
cd build_release
ls -lh lib/*/*
- name: Build wheel
shell: bash
run: |
echo $PWD
ls -lh ./*
export KALDIFEAT_CMAKE_ARGS="-DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCUDNN_INCLUDE_PATH=d:/a/kaldifeat/kaldifeat/cudnn-for-windows/cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive/include -DCUDNN_LIBRARY_PATH=d:/a/kaldifeat/kaldifeat/cudnn-for-windows/cudnn-windows-x86_64-8.4.1.50_cuda11.6-archive/lib/cudnn.lib"
python3 setup.py bdist_wheel
ls -lh dist/
pip install ./dist/*.whl
python3 -c "import kaldifeat; print(kaldifeat.__version__)"
- name: Upload Wheel
uses: actions/upload-artifact@v2
with:
name: python-${{ matrix.python-version }}-${{ matrix.os }}-cuda-${{ matrix.cuda }}
path: dist/*.whl
- name: Build tests
shell: bash
run: |
cd build_release
cmake --build . --target ALL_BUILD --config Release
ls -lh bin/*/*
ctest -C Release --verbose --output-on-failure

1
.gitignore vendored
View File

@ -5,3 +5,4 @@ dist/
__pycache__/
test-1hour.wav
path.sh
torch_version.py

21
CMakeLists.txt
View File

@ -6,7 +6,8 @@ project(kaldifeat)
# remember to change the version in
# scripts/conda/kaldifeat/meta.yaml
set(kaldifeat_VERSION "1.14")
# scripts/conda-cpu/kaldifeat/meta.yaml
set(kaldifeat_VERSION "1.16")
set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/lib")
set(CMAKE_LIBRARY_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/lib")
@ -51,5 +52,23 @@ if(kaldifeat_BUILD_TESTS)
enable_testing()
endif()
# TORCH_VERSION is defined in cmake/torch.cmake
configure_file(
${CMAKE_SOURCE_DIR}/kaldifeat/python/kaldifeat/torch_version.py.in
${CMAKE_SOURCE_DIR}/kaldifeat/python/kaldifeat/torch_version.py @ONLY
)
if(WIN32)
# disable various warnings for MSVC
# 4624: destructor was implicitly defined as deleted because a base class destructor is inaccessible or deleted
set(disabled_warnings
/wd4624
)
message(STATUS "Disabled warnings: ${disabled_warnings}")
foreach(w IN LISTS disabled_warnings)
string(APPEND CMAKE_CXX_FLAGS " ${w} ")
endforeach()
endif()
include_directories(${CMAKE_SOURCE_DIR})
add_subdirectory(kaldifeat)

78
README.md
View File

@ -1,12 +1,16 @@
# kaldifeat
<div align="center">
<img src="/doc/source/images/os.svg">
<img src="/doc/source/images/os-green.svg">
<img src="/doc/source/images/python_ge_3.6-blue.svg">
<img src="/doc/source/images/pytorch_ge_1.5.0-green.svg">
<img src="/doc/source/images/cuda_ge_10.1-orange.svg">
</div>
[![Documentation Status](https://github.com/csukuangfj/kaldifeat/actions/workflows/build-doc.yml/badge.svg)](https://csukuangfj.github.io/kaldifeat/)
**Documentation**: <https://csukuangfj.github.io/kaldifeat>
<sub>
<table>
<tr>
@ -268,67 +272,15 @@ See <https://github.com/k2-fsa/k2/blob/v2.0-pre/k2/torch/csrc/features.cu>.
See <https://github.com/lhotse-speech/lhotse/blob/master/lhotse/features/kaldifeat.py>.
## sherpa
[sherpa](https://github.com/k2-fsa/sherpa) uses kaldifeat for streaming feature
extraction.
See <https://github.com/k2-fsa/sherpa/blob/master/sherpa/bin/pruned_stateless_emformer_rnnt2/decode.py>
# Installation
## From conda (Only for Linux + CUDA)
Supported versions of Python, PyTorch, and CUDA toolkit are listed below:
[![Supported Python versions](/doc/source/images/python-3.6_3.7_3.8-blue.svg)](/doc/source/images/python-3.6_3.7_3.8-blue.svg)
[![Supported PyTorch versions](/doc/source/images/pytorch-1.6.0_1.7.0_1.7.1_1.8.0_1.8.1_1.9.0-green.svg)](/doc/source/images/pytorch-1.6.0_1.7.0_1.7.1_1.8.0_1.8.1_1.9.0-green.svg)
[![Supported CUDA versions](/doc/source/images/cuda-10.1_10.2_11.0_11.1-orange.svg)](/doc/source/images/cuda-10.1_10.2_11.0_11.1-orange.svg)
```bash
conda install -c kaldifeat -c pytorch -c conda-forge kaldifeat python=3.8 cudatoolkit=11.1 pytorch=1.8.1
```
You can select the supported Python version, CUDA toolkit version and PyTorch version as you wish.
**Note:** If you want a CPU only version or want to install `kaldifeat` on other operating systems,
e.g., macOS, please use `pip install` or compile `kaldifeat` from source.
## From PyPi with pip
You need to install PyTorch and CMake first.
CMake 3.11 is known to work. Other CMake versions may also work.
PyTorch 1.6.0 and above are known to work. Other PyTorch versions may also work.
```bash
pip install -v kaldifeat
```
## From source
The following are the commands to compile `kaldifeat` from source.
We assume that you have installed `CMake` and PyTorch.
CMake 3.11 is known to work. Other CMake versions may also work.
PyTorch 1.6.0 and above are known to work. Other PyTorch versions may also work.
```bash
mkdir /some/path
git clone https://github.com/csukuangfj/kaldifeat.git
cd kaldifeat
python setup.py install
```
To test whether `kaldifeat` was installed successfully, you can run:
```bash
python3 -c "import kaldifeat; print(kaldifeat.__version__)"
```
## FAQs
### How to install a CUDA version
There are two approaches:
- (1) Install using `conda`. It always installs a CUDA version of kaldifeat.
- (2) Install a CUDA version of PyTorch and then install kaldifeat from source
or use `pip install kaldifeat`.
### How to install a CPU-only version
You have to first install a CPU-only version of PyTorch and then install kaldifeat
from source or use `pip install kaldifeat`.
Refer to
<https://csukuangfj.github.io/kaldifeat>
for installation.

8
cmake/cmake_extension.py
View File

@ -128,3 +128,11 @@ class BuildExtension(build_ext):
for so in lib_so:
print(f"Copying {so} to {self.build_lib}/")
shutil.copy(f"{so}", f"{self.build_lib}/")
print(
f"Copying {kaldifeat_dir}/kaldifeat/python/kaldifeat/torch_version.py to {self.build_lib}/kaldifeat" # noqa
)
shutil.copy(
f"{kaldifeat_dir}/kaldifeat/python/kaldifeat/torch_version.py",
f"{self.build_lib}/kaldifeat",
)

7
cmake/pybind11.cmake
View File

@ -8,12 +8,9 @@ function(download_pybind11)
include(FetchContent)
set(pybind11_URL "https://github.com/pybind/pybind11/archive/v2.6.0.tar.gz")
set(pybind11_HASH "SHA256=90b705137b69ee3b5fc655eaca66d0dc9862ea1759226f7ccd3098425ae69571")
set(pybind11_URL "https://github.com/pybind/pybind11/archive/v2.9.2.tar.gz")
set(pybind11_HASH "SHA256=6bd528c4dbe2276635dc787b6b1f2e5316cf6b49ee3e150264e455a0d68d19c1")
set(double_quotes "\"")
set(dollar "\$")
set(semicolon "\;")
FetchContent_Declare(pybind11
URL ${pybind11_URL}
URL_HASH ${pybind11_HASH}

0
doc/source/_static/.gitkeep Normal file
View File

72
doc/source/code/test_fbank.py
View File

@ -1,72 +0,0 @@
#!/usr/bin/env python3
# Copyright 2021 Xiaomi Corporation (authors: Fangjun Kuang)
import numpy as np
import soundfile as sf
import torch
import kaldifeat
def read_wave(filename) -> torch.Tensor:
"""Read a wave file and return it as a 1-D tensor.
Note:
You don't need to scale it to [-32768, 32767].
We use scaling here to follow the approach in Kaldi.
Args:
filename:
Filename of a sound file.
Returns:
Return a 1-D tensor containing audio samples.
"""
with sf.SoundFile(filename) as sf_desc:
sampling_rate = sf_desc.samplerate
assert sampling_rate == 16000
data = sf_desc.read(dtype=np.float32, always_2d=False)
data *= 32768
return torch.from_numpy(data)
def test_fbank():
device = torch.device("cpu")
if torch.cuda.is_available():
device = torch.device("cuda", 0)
wave0 = read_wave("test_data/test.wav")
wave1 = read_wave("test_data/test2.wav")
wave0 = wave0.to(device)
wave1 = wave1.to(device)
opts = kaldifeat.FbankOptions()
opts.frame_opts.dither = 0
opts.device = device
fbank = kaldifeat.Fbank(opts)
# We can compute fbank features in batches
features = fbank([wave0, wave1])
assert isinstance(features, list), f"{type(features)}"
assert len(features) == 2
# We can also compute fbank features for a single wave
features0 = fbank(wave0)
features1 = fbank(wave1)
assert torch.allclose(features[0], features0)
assert torch.allclose(features[1], features1)
# To compute fbank features for only a specified frame
audio_frames = fbank.convert_samples_to_frames(wave0)
feature_frame_1 = fbank.compute(audio_frames[1])
feature_frame_10 = fbank.compute(audio_frames[10])
assert torch.allclose(features0[1], feature_frame_1)
assert torch.allclose(features0[10], feature_frame_10)
if __name__ == "__main__":
test_fbank()

34
doc/source/conf.py
View File

@ -59,7 +59,7 @@ templates_path = ["_templates"]
# List of patterns, relative to source directory, that match files and
# directories to ignore when looking for source files.
# This pattern also affects html_static_path and html_extra_path.
exclude_patterns = []
exclude_patterns = ["images/*.md"]
source_suffix = {
".rst": "restructuredtext",
@ -102,3 +102,35 @@ html_theme_options = {
"prev_next_buttons_location": "bottom",
"style_external_links": True,
}
rst_epilog = """
.. _kaldifeat: https://github.com/csukuangfj/kaldifeat
.. _Kaldi: https://github.com/kaldi-asr/kaldi
.. _PyTorch: https://pytorch.org/
.. _kaldifeat.Fbank: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/kaldifeat/fbank.py#L10
.. _kaldifeat.Mfcc: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/kaldifeat/mfcc.py#L10
.. _kaldifeat.Plp: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/kaldifeat/plp.py#L10
.. _kaldifeat.Spectrogram: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/kaldifeat/spectrogram.py#L9
.. _kaldifeat.OnlineFbank: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/kaldifeat/fbank.py#L16
.. _kaldifeat.OnlineMfcc: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/kaldifeat/mfcc.py#L16
.. _kaldifeat.OnlinePlp: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/kaldifeat/plp.py#L16
.. _compute-fbank-feats: https://github.com/kaldi-asr/kaldi/blob/master/src/featbin/compute-fbank-feats.cc
.. _compute-mfcc-feats: https://github.com/kaldi-asr/kaldi/blob/master/src/featbin/compute-mfcc-feats.cc
.. _compute-plp-feats: https://github.com/kaldi-asr/kaldi/blob/master/src/featbin/compute-plp-feats.cc
.. _compute-spectrogram-feats: https://github.com/kaldi-asr/kaldi/blob/master/src/featbin/compute-spectrogram-feats.cc
.. _kaldi::OnlineFbank: https://github.com/kaldi-asr/kaldi/blob/master/src/feat/online-feature.h#L160
.. _kaldi::OnlineMfcc: https://github.com/kaldi-asr/kaldi/blob/master/src/feat/online-feature.h#L158
.. _kaldi::OnlinePlp: https://github.com/kaldi-asr/kaldi/blob/master/src/feat/online-feature.h#L159
.. _kaldifeat.FbankOptions: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/csrc/feature-fbank.h#L19
.. _kaldi::FbankOptions: https://github.com/kaldi-asr/kaldi/blob/master/src/feat/feature-fbank.h#L41
.. _kaldifeat.MfccOptions: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/csrc/feature-mfcc.h#L22
.. _kaldi::MfccOptions: https://github.com/kaldi-asr/kaldi/blob/master/src/feat/feature-mfcc.h#L38
.. _kaldifeat.PlpOptions: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/csrc/feature-plp.h#L24
.. _kaldi::PlpOptions: https://github.com/kaldi-asr/kaldi/blob/master/src/feat/feature-plp.h#L42
.. _kaldifeat.SpectrogramOptions: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/csrc/feature-spectrogram.h#L18
.. _kaldi::SpectrogramOptions: https://github.com/kaldi-asr/kaldi/blob/master/src/feat/feature-spectrogram.h#L38
.. _kaldifeat.FrameExtractionOptions: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/csrc/feature-window.h#L30
.. _kaldi::FrameExtractionOptions: https://github.com/kaldi-asr/kaldi/blob/master/src/feat/feature-window.h#L35
.. _kaldifeat.MelBanksOptions: https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/csrc/mel-computations.h#L17
.. _kaldi::MelBanksOptions: https://github.com/kaldi-asr/kaldi/blob/master/src/feat/mel-computations.h#L43
"""

8
doc/source/images/README.md Normal file
View File

@ -0,0 +1,8 @@
## File description
<https://shields.io/> is used to create the following files:
- ./os.svg
- ./python_ge_3.6-blue.svg
- ./cuda_ge_10.1-orange.svg
- ./pytorch_ge_1.5.0-green.svg

1
doc/source/images/os-green.svg Normal file
View File

@ -0,0 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" width="176" height="20" role="img" aria-label="os: Linux | macOS | Windows"><title>os: Linux | macOS | Windows</title><linearGradient id="s" x2="0" y2="100%"><stop offset="0" stop-color="#bbb" stop-opacity=".1"/><stop offset="1" stop-opacity=".1"/></linearGradient><clipPath id="r"><rect width="176" height="20" rx="3" fill="#fff"/></clipPath><g clip-path="url(#r)"><rect width="23" height="20" fill="#555"/><rect x="23" width="153" height="20" fill="#97ca00"/><rect width="176" height="20" fill="url(#s)"/></g><g fill="#fff" text-anchor="middle" font-family="Verdana,Geneva,DejaVu Sans,sans-serif" text-rendering="geometricPrecision" font-size="110"><text aria-hidden="true" x="125" y="150" fill="#010101" fill-opacity=".3" transform="scale(.1)" textLength="130">os</text><text x="125" y="140" transform="scale(.1)" fill="#fff" textLength="130">os</text><text aria-hidden="true" x="985" y="150" fill="#010101" fill-opacity=".3" transform="scale(.1)" textLength="1430">Linux | macOS | Windows</text><text x="985" y="140" transform="scale(.1)" fill="#fff" textLength="1430">Linux | macOS | Windows</text></g></svg>
Side by Side

After

Width:  |  Height:  |  Size: 1.2 KiB

1
doc/source/images/os.svg
View File

@ -1 +0,0 @@
<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" width="118" height="20" role="img" aria-label="OS: Linux | macOS"><title>OS: Linux | macOS</title><linearGradient id="s" x2="0" y2="100%"><stop offset="0" stop-color="#bbb" stop-opacity=".1"/><stop offset="1" stop-opacity=".1"/></linearGradient><clipPath id="r"><rect width="118" height="20" rx="3" fill="#fff"/></clipPath><g clip-path="url(#r)"><rect width="27" height="20" fill="#555"/><rect x="27" width="91" height="20" fill="#4c1"/><rect width="118" height="20" fill="url(#s)"/></g><g fill="#fff" text-anchor="middle" font-family="Verdana,Geneva,DejaVu Sans,sans-serif" text-rendering="geometricPrecision" font-size="110"><text aria-hidden="true" x="145" y="150" fill="#010101" fill-opacity=".3" transform="scale(.1)" textLength="170">OS</text><text x="145" y="140" transform="scale(.1)" fill="#fff" textLength="170">OS</text><text aria-hidden="true" x="715" y="150" fill="#010101" fill-opacity=".3" transform="scale(.1)" textLength="810">Linux | macOS</text><text x="715" y="140" transform="scale(.1)" fill="#fff" textLength="810">Linux | macOS</text></g></svg>
Side by Side

Before

Width:  |  Height:  |  Size: 1.1 KiB

14
doc/source/index.rst
View File

@ -6,19 +6,11 @@
kaldifeat
=========
`kaldifeat <https://github.com/csukuangfj/kaldifeat>`_ implements
feature extraction algorithms **compatible** with kaldi using PyTorch, supporting CUDA
as well as autograd.
Currently, only fbank features are supported.
It can produce the same feature output as ``compute-fbank-feats`` (from kaldi)
when given the same options.
.. toctree::
:maxdepth: 2
:caption: Contents:
:caption: Contents
intro
installation
usage
usage/index

141
doc/source/installation.rst
View File

@ -1,19 +1,38 @@
Installation
============
- |os_types|
- |python_versions|
- |pytorch_versions|
- |cuda_versions|
.. caution::
`kaldifeat`_ depends on `PyTorch`_. `PyTorch`_ >= 1.5.0 is known to work.
Please first install `PyTorch`_ before you install `kaldifeat`_.
.. hint::
To install a CPU version of `kaldifeat`_, please install a CPU version
of `PyTorch`_.
To install a CUDA version of `kaldifeat`_, please install a CUDA version
of `PyTorch`_. CUDA >= 10.1 is known to work.
.. _from source:
Install kaldifeat from source
-----------------------------
You have to install ``cmake`` and ``PyTorch`` first.
You have to install ``cmake`` and `PyTorch`_ first.
- ``cmake`` 3.11 is known to work. Other CMake versions may also work.
- ``PyTorch`` 1.8.1 is known to work. Other PyTorch versions may also work.
- `PyTorch`_ >= 1.5.0 is known to work. Other PyTorch versions may also work.
- Python >= 3.6
The commands to install ``kaldifeat`` from source are:
The commands to install `kaldifeat`_ from source are:
.. code-block:: bash
@ -21,7 +40,7 @@ The commands to install ``kaldifeat`` from source are:
cd kaldifeat
python3 setup.py install
To test that you have installed ``kaldifeat`` successfully, please run:
To test that you have installed `kaldifeat`_ successfully, please run:
.. code-block:: bash
@ -29,26 +48,120 @@ To test that you have installed ``kaldifeat`` successfully, please run:
It should print the version, e.g., ``1.0``.
.. _from PyPI:
Install kaldifeat from PyPI
---------------------------
The pre-built ``kaldifeat`` hosted on PyPI uses PyTorch 1.8.1.
If you install ``kaldifeat`` using pip, it will replace your locally
installed PyTorch automatically with PyTorch 1.8.1.
If you don't want this happen, please `Install kaldifeat from source`_.
The command to install ``kaldifeat`` from PyPI is:
The command to install `kaldifeat`_ from PyPI is:
.. code-block:: bash
pip install kaldifeat
pip install --verbose kaldifeat
To test that you have installed ``kaldifeat`` successfully, please run:
To test that you have installed `kaldifeat`_ successfully, please run:
.. code-block:: bash
python3 -c "import kaldifeat; print(kaldifeat.__version__)"
It should print the version, e.g., ``1.0``.
Install kaldifeat from conda (Only for Linux)
---------------------------------------------
.. hint::
Installation using ``conda`` supports only Linux. For macOS and Windows,
please use either :ref:`from source` or :ref:`from PyPI`.
The command to install `kaldifeat` using ``conda`` is
.. code-block:: bash
conda install -c kaldifeat -c pytorch -c conda-forge kaldifeat python=3.8 cudatoolkit=11.1 pytorch=1.8.1
You can select the supported Python version, CUDA toolkit version and `PyTorch`_ version as you wish.
To install a CPU version of `kaldifeat`, use:
.. code-block:: bash
conda install -c kaldifeat -c pytorch cpuonly kaldifeat python=3.8 pytorch=1.8.1
.. caution::
If you encounter issues about missing GLIBC after installing `kaldifeat`_
with ``conda``, please consider :ref:`from source` or :ref:`from PyPI`.
The reason is that the package was built using Ubuntu 18.04 and your system's
GLIBC is older.
.. |os_types| image:: ./images/os-green.svg
:alt: Supported operating systems
.. |python_versions| image:: ./images/python_ge_3.6-blue.svg
:alt: Supported python versions
.. |cuda_versions| image:: ./images/cuda_ge_10.1-orange.svg
:alt: Supported cuda versions
.. |pytorch_versions| image:: ./images/pytorch_ge_1.5.0-green.svg
:alt: Supported pytorch versions
To test that you have installed `kaldifeat`_ successfully, please run:
.. code-block:: bash
python3 -c "import kaldifeat; print(kaldifeat.__version__)"
It should print the version, e.g., ``1.0``.
FAQs
----
How to install a CUDA version of kaldifeat
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
You need to first install a CUDA version of `PyTorch`_ and then install `kaldifeat`_.
.. note::
You can use a CUDA version of `kaldifeat`_ on machines with no GPUs.
How to install a CPU version of kaldifeat
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
You need to first install a CPU version of `PyTorch`_ and then install `kaldifeat`_.
How to fix `Caffe2: Cannot find cuDNN library`
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
.. code-block::
Your installed Caffe2 version uses cuDNN but I cannot find the cuDNN
libraries. Please set the proper cuDNN prefixes and / or install cuDNN.
You will have such an error when you want to install a CUDA version of `kaldifeat`_
by ``pip install kaldifeat`` or from source.
You need to first install cuDNN. Assume you have installed cuDNN to the
path ``/path/to/cudnn``. You can fix the error by using ``one`` of the following
commands.
(1) Fix for installation using ``pip install``
.. code-block:: bash
export KALDIFEAT_CMAKE_ARGS="-DCMAKE_BUILD_TYPE=Release -DCUDNN_LIBRARY_PATH=/path/to/cudnn/lib/libcudnn.so -DCUDNN_INCLUDE_PATH=/path/to/cudnn/include"
pip install --verbose kaldifeat
(2) Fix for installation from source
.. code-block:: bash
mkdir /some/path
git clone https://github.com/csukuangfj/kaldifeat.git
cd kaldifeat
export KALDIFEAT_CMAKE_ARGS="-DCMAKE_BUILD_TYPE=Release -DCUDNN_LIBRARY_PATH=/path/to/cudnn/lib/libcudnn.so -DCUDNN_INCLUDE_PATH=/path/to/cudnn/include"
python setup.py install

103
doc/source/intro.rst Normal file
View File

@ -0,0 +1,103 @@
Introduction
============
`kaldifeat`_ implements
speech feature extraction algorithms **compatible** with `Kaldi`_ using `PyTorch`_,
supporting CUDA as well as autograd.
`kaldifeat`_ has the following features:
- Fully compatible with `Kaldi`_
.. note::
The underlying C++ code is copied & modified from `Kaldi`_ directly.
It is rewritten with `PyTorch` C++ APIs.
- Provide not only ``C++ APIs`` but also ``Python APIs``
.. note::
You can access `kaldifeat`_ from ``Python``.
- Support autograd
- Support ``CUDA`` and ``CPU``
.. note::
You can use CUDA for feature extraction.
- Support ``online`` (i.e., ``streaming``) and ``offline`` (i.e., ``non-streaming``)
feature extraction
- Support chunk-based processing
.. note::
This is especially usefull if you want to process audios of several
hours long, which may cause OOM if you send them for computation at once.
With chunk-based processing, you can process audios of arbirtray length.
- Support batch processing
.. note::
With `kaldifeat`_ you can extract features for a batch of audios
.. see https://sublime-and-sphinx-guide.readthedocs.io/en/latest/tables.html
Currently implemented speech features and their counterparts in `Kaldi`_ are
listed in the following table.
.. list-table:: Supported speech features
:widths: 50 50
:header-rows: 1
* - Supported speech features
- Counterpart in `Kaldi`_
* - `kaldifeat.Fbank`_
- `compute-fbank-feats`_
* - `kaldifeat.Mfcc`_
- `compute-mfcc-feats`_
* - `kaldifeat.Plp`_
- `compute-plp-feats`_
* - `kaldifeat.Spectrogram`_
- `compute-spectrogram-feats`_
* - `kaldifeat.OnlineFbank`_
- `kaldi::OnlineFbank`_
* - `kaldifeat.OnlineMfcc`_
- `kaldi::OnlineMfcc`_
* - `kaldifeat.OnlinePlp`_
- `kaldi::OnlinePlp`_
Each feature computer needs an option. The following table lists the options
for each computer and the corresponding options in `Kaldi`_.
.. hint::
Note that we reuse the parameter names from `Kaldi`_.
Also, both online feature computers and offline feature computers share the
same option.
.. list-table:: Feature computer options
:widths: 50 50
:header-rows: 1
* - Options in `kaldifeat`_
- Corresponding options in `Kaldi`_
* - `kaldifeat.FbankOptions`_
- `kaldi::FbankOptions`_
* - `kaldifeat.MfccOptions`_
- `kaldi::MfccOptions`_
* - `kaldifeat.PlpOptions`_
- `kaldi::PlpOptions`_
* - `kaldifeat.SpectrogramOptions`_
- `kaldi::SpectrogramOptions`_
* - `kaldifeat.FrameExtractionOptions`_
- `kaldi::FrameExtractionOptions`_
* - `kaldifeat.MelBanksOptions`_
- `kaldi::MelBanksOptions`_
Read more to learn how to install `kaldifeat`_ and how to use each feature
computer.

212
doc/source/usage.rst
View File

@ -1,212 +0,0 @@
Usage
=====
Let us first see the help message of kaldi's ``compute-fbank-feats``:
.. code-block:: bash
$ compute-fbank-feats
Create Mel-filter bank (FBANK) feature files.
Usage: compute-fbank-feats [options...] <wav-rspecifier> <feats-wspecifier>
Options:
--allow-downsample : If true, allow the input waveform to have a higher frequency than the specified --sample-frequency (and we'll downsample). (bool, default = false)
--allow-upsample : If true, allow the input waveform to have a lower frequency than the specified --sample-frequency (and we'll upsample). (bool, default = false)
--blackman-coeff : Constant coefficient for generalized Blackman window. (float, default = 0.42)
--channel : Channel to extract (-1 -> expect mono, 0 -> left, 1 -> right) (int, default = -1)
--debug-mel : Print out debugging information for mel bin computation (bool, default = false)
--dither : Dithering constant (0.0 means no dither). If you turn this off, you should set the --energy-floor option, e.g. to 1.0 or 0.1 (float, default = 1)
--energy-floor : Floor on energy (absolute, not relative) in FBANK computation. Only makes a difference if --use-energy=true; only necessary if --dither=0.0. Suggested values: 0.1 or 1.0 (float, default = 0)
--frame-length : Frame length in milliseconds (float, default = 25)
--frame-shift : Frame shift in milliseconds (float, default = 10)
--high-freq : High cutoff frequency for mel bins (if <= 0, offset from Nyquist) (float, default = 0)
--htk-compat : If true, put energy last. Warning: not sufficient to get HTK compatible features (need to change other parameters). (bool, default = false)
--low-freq : Low cutoff frequency for mel bins (float, default = 20)
--max-feature-vectors : Memory optimization. If larger than 0, periodically remove feature vectors so that only this number of the latest feature vectors is retained. (int, default = -1)
--min-duration : Minimum duration of segments to process (in seconds). (float, default = 0)
--num-mel-bins : Number of triangular mel-frequency bins (int, default = 23)
--output-format : Format of the output files [kaldi, htk] (string, default = "kaldi")
--preemphasis-coefficient : Coefficient for use in signal preemphasis (float, default = 0.97)
--raw-energy : If true, compute energy before preemphasis and windowing (bool, default = true)
--remove-dc-offset : Subtract mean from waveform on each frame (bool, default = true)
--round-to-power-of-two : If true, round window size to power of two by zero-padding input to FFT. (bool, default = true)
--sample-frequency : Waveform data sample frequency (must match the waveform file, if specified there) (float, default = 16000)
--snip-edges : If true, end effects will be handled by outputting only frames that completely fit in the file, and the number of frames depends on the frame-length. If false, the number of frames depends only on the frame-shift, and we reflect the data at the ends. (bool, default = true)
--subtract-mean : Subtract mean of each feature file [CMS]; not recommended to do it this way. (bool, default = false)
--use-energy : Add an extra dimension with energy to the FBANK output. (bool, default = false)
--use-log-fbank : If true, produce log-filterbank, else produce linear. (bool, default = true)
--use-power : If true, use power, else use magnitude. (bool, default = true)
--utt2spk : Utterance to speaker-id map (if doing VTLN and you have warps per speaker) (string, default = "")
--vtln-high : High inflection point in piecewise linear VTLN warping function (if negative, offset from high-mel-freq (float, default = -500)
--vtln-low : Low inflection point in piecewise linear VTLN warping function (float, default = 100)
--vtln-map : Map from utterance or speaker-id to vtln warp factor (rspecifier) (string, default = "")
--vtln-warp : Vtln warp factor (only applicable if vtln-map not specified) (float, default = 1)
--window-type : Type of window ("hamming"|"hanning"|"povey"|"rectangular"|"sine"|"blackmann") (string, default = "povey")
--write-utt2dur : Wspecifier to write duration of each utterance in seconds, e.g. 'ark,t:utt2dur'. (string, default = "")
Standard options:
--config : Configuration file to read (this option may be repeated) (string, default = "")
--help : Print out usage message (bool, default = false)
--print-args : Print the command line arguments (to stderr) (bool, default = true)
--verbose : Verbose level (higher->more logging) (int, default = 0)
FbankOptions
------------
``kaldifeat`` reuses the same options from kaldi's ``compute-fbank-feats``.
The following shows the default values of ``kaldifeat.FbankOptions``:
.. code-block:: python
>>> import kaldifeat
>>> fbank_opts = kaldifeat.FbankOptions()
>>> print(fbank_opts)
frame_opts:
samp_freq: 16000
frame_shift_ms: 10
frame_length_ms: 25
dither: 1
preemph_coeff: 0.97
remove_dc_offset: 1
window_type: povey
round_to_power_of_two: 1
blackman_coeff: 0.42
snip_edges: 1
mel_opts:
num_bins: 23
low_freq: 20
high_freq: 0
vtln_low: 100
vtln_high: -500
debug_mel: 0
htk_mode: 0
use_energy: 0
energy_floor: 0
raw_energy: 1
htk_compat: 0
use_log_fbank: 1
use_power: 1
device: cpu
It consists of three parts:
- ``frame_opts``
Options in this part are accessed by ``frame_opts.xxx``. That is, to access
the sample rate, you use:
.. code-block:: python
>>> fbank_opts = kaldifeat.FbankOptions()
>>> print(fbank_opts.frame_opts.samp_freq)
16000.0
- ``mel_opts``
Options in this part are accessed by ``mel_opts.xxx``. That is, to access
the number of mel bins, you use:
.. code-block:: python
>>> fbank_opts = kaldifeat.FbankOptions()
>>> print(fbank_opts.mel_opts.num_bins)
23
- fbank related
Options in this part are accessed directly. That is, to access the device
field, you use:
.. code-block::
>>> print(fbank_opts.device)
cpu
>>> fbank_opts.device = 'cuda:0'
>>> print(fbank_opts.device)
cuda:0
>>> import torch
>>> fbank_opts.device = torch.device('cuda', 0)
>>> print(fbank_opts.device)
cuda:0
To change the sample rate to 8000, you can use:
.. code-block:: python
>>> fbank_opts = kaldifeat.FbankOptions()
>>> print(fbank_opts.frame_opts.samp_freq)
16000.0
>>> fbank_opts.frame_opts.samp_freq = 8000
>>> print(fbank_opts.frame_opts.samp_freq)
8000.0
To change ``snip_edges`` to ``False``, you can use:
.. code-block:: python
>>> fbank_opts.frame_opts.snip_edges = False
>>> print(fbank_opts.frame_opts.snip_edges)
False
To change number of mel bins to 80, you can use:
.. code-block:: python
>>> print(fbank_opts.mel_opts.num_bins)
23
>>> fbank_opts.mel_opts.num_bins = 80
>>> print(fbank_opts.mel_opts.num_bins)
80
To change the device to ``cuda``, you can use:
Fbank
-----
The following shows how to use ``kaldifeat.Fbank`` to compute
the fbank features of sound files.
First, let us generate two sound files using ``sox``:
.. code-block:: bash
# generate a wav of two seconds, containing a sine-wave
# swept from 300 Hz to 3300 Hz
sox -n -r 16000 -b 16 test.wav synth 1.2 sine 300-3300
# another sound file with 0.5 seconds
sox -n -r 16000 -b 16 test2.wav synth 0.5 sine 300-3300
.. hint::
You can find the above two files by visiting the following two links:
- `test.wav <https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/tests/test_data/test.wav>`_
- `test2.wav <https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/tests/test_data/test2.wav>`_
The `following code <https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/tests/test_fbank.py>`_
shows the usage of ``kaldifeat.Fbank``.
It shows:
- How to read a sound file. Note that audio samples are scaled to the range [-32768, 32768].
The intention is to produce the same output as kaldi. You don't need to scale it if
you don't care about the compatibility with kaldi
- ``kaldifeat.Fbank`` supports CUDA as well as CPU
- ``kaldifeat.Fbank`` supports processing sound file in a batch as well as accepting
a single sound file
.. literalinclude:: ./code/test_fbank.py
:caption: Demo of ``kaldifeat.Fbank``
:language: python

46
doc/source/usage/code/compute-fbank-feats-help.txt Normal file
View File

@ -0,0 +1,46 @@
compute-fbank-feats
Create Mel-filter bank (FBANK) feature files.
Usage: compute-fbank-feats [options...] <wav-rspecifier> <feats-wspecifier>
Options:
--allow-downsample : If true, allow the input waveform to have a higher frequency than the specified --sample-frequency (and we'll downsample). (bool, default = false)
--allow-upsample : If true, allow the input waveform to have a lower frequency than the specified --sample-frequency (and we'll upsample). (bool, default = false)
--blackman-coeff : Constant coefficient for generalized Blackman window. (float, default = 0.42)
--channel : Channel to extract (-1 -> expect mono, 0 -> left, 1 -> right) (int, default = -1)
--debug-mel : Print out debugging information for mel bin computation (bool, default = false)
--dither : Dithering constant (0.0 means no dither). If you turn this off, you should set the --energy-floor option, e.g. to 1.0 or 0.1 (float, default = 1)
--energy-floor : Floor on energy (absolute, not relative) in FBANK computation. Only makes a difference if --use-energy=true; only necessary if --dither=0.0. Suggested values: 0.1 or 1.0 (float, default = 0)
--frame-length : Frame length in milliseconds (float, default = 25)
--frame-shift : Frame shift in milliseconds (float, default = 10)
--high-freq : High cutoff frequency for mel bins (if <= 0, offset from Nyquist) (float, default = 0)
--htk-compat : If true, put energy last. Warning: not sufficient to get HTK compatible features (need to change other parameters). (bool, default = false)
--low-freq : Low cutoff frequency for mel bins (float, default = 20)
--max-feature-vectors : Memory optimization. If larger than 0, periodically remove feature vectors so that only this number of the latest feature vectors is retained. (int, default = -1)
--min-duration : Minimum duration of segments to process (in seconds). (float, default = 0)
--num-mel-bins : Number of triangular mel-frequency bins (int, default = 23)
--output-format : Format of the output files [kaldi, htk] (string, default = "kaldi")
--preemphasis-coefficient : Coefficient for use in signal preemphasis (float, default = 0.97)
--raw-energy : If true, compute energy before preemphasis and windowing (bool, default = true)
--remove-dc-offset : Subtract mean from waveform on each frame (bool, default = true)
--round-to-power-of-two : If true, round window size to power of two by zero-padding input to FFT. (bool, default = true)
--sample-frequency : Waveform data sample frequency (must match the waveform file, if specified there) (float, default = 16000)
--snip-edges : If true, end effects will be handled by outputting only frames that completely fit in the file, and the number of frames depends on the frame-length. If false, the number of frames depends only on the frame-shift, and we reflect the data at the ends. (bool, default = true)
--subtract-mean : Subtract mean of each feature file [CMS]; not recommended to do it this way. (bool, default = false)
--use-energy : Add an extra dimension with energy to the FBANK output. (bool, default = false)
--use-log-fbank : If true, produce log-filterbank, else produce linear. (bool, default = true)
--use-power : If true, use power, else use magnitude. (bool, default = true)
--utt2spk : Utterance to speaker-id map (if doing VTLN and you have warps per speaker) (string, default = "")
--vtln-high : High inflection point in piecewise linear VTLN warping function (if negative, offset from high-mel-freq (float, default = -500)
--vtln-low : Low inflection point in piecewise linear VTLN warping function (float, default = 100)
--vtln-map : Map from utterance or speaker-id to vtln warp factor (rspecifier) (string, default = "")
--vtln-warp : Vtln warp factor (only applicable if vtln-map not specified) (float, default = 1)
--window-type : Type of window ("hamming"|"hanning"|"povey"|"rectangular"|"sine"|"blackmann") (string, default = "povey")
--write-utt2dur : Wspecifier to write duration of each utterance in seconds, e.g. 'ark,t:utt2dur'. (string, default = "")
Standard options:
--config : Configuration file to read (this option may be repeated) (string, default = "")
--help : Print out usage message (bool, default = false)
--print-args : Print the command line arguments (to stderr) (bool, default = true)
--verbose : Verbose level (higher->more logging) (int, default = 0)

65
doc/source/usage/code/fbank_options-1.txt Normal file
View File

@ -0,0 +1,65 @@
$ python3
Python 3.8.0 (default, Oct 28 2019, 16:14:01)
[GCC 8.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import kaldifeat
>>> opts = kaldifeat.FbankOptions()
>>> print(opts)
frame_opts:
samp_freq: 16000
frame_shift_ms: 10
frame_length_ms: 25
dither: 1
preemph_coeff: 0.97
remove_dc_offset: 1
window_type: povey
round_to_power_of_two: 1
blackman_coeff: 0.42
snip_edges: 1
max_feature_vectors: -1
mel_opts:
num_bins: 23
low_freq: 20
high_freq: 0
vtln_low: 100
vtln_high: -500
debug_mel: 0
htk_mode: 0
use_energy: 0
energy_floor: 0
raw_energy: 1
htk_compat: 0
use_log_fbank: 1
use_power: 1
device: cpu
>>> print(opts.dither)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
AttributeError: '_kaldifeat.FbankOptions' object has no attribute 'dither'
>>>
>>> print(opts.frame_opts.dither)
1.0
>>> opts.frame_opts.dither = 0 # disable dither
>>> print(opts.frame_opts.dither)
0.0
>>> import torch
>>> print(opts.device)
cpu
>>> opts.device = 'cuda:0'
>>> print(opts.device)
cuda:0
>>> opts.device = torch.device('cuda', 1)
>>> print(opts.device)
cuda:1
>>> opts.device = 'cpu'
>>> print(opts.device)
cpu
>>> print(opts.mel_opts.num_bins)
23
>>> opts.mel_opts.num_bins = 80
>>> print(opts.mel_opts.num_bins)
80

1
doc/source/usage/code/test_fbank_options.py Symbolic link
View File

@ -0,0 +1 @@
../../../../kaldifeat/python/tests/test_fbank_options.py

3
doc/source/usage/fbank.rst Normal file
View File

@ -0,0 +1,3 @@
kaldifeat.Fbank
===============

52
doc/source/usage/fbank_options.rst Normal file
View File

@ -0,0 +1,52 @@
kaldifeat.FbankOptions
======================
If you want to construct an instance of `kaldifeat.Fbank`_ or
`kaldifeat.OnlineFbank`_, you have to provide an instance of
`kaldifeat.FbankOptions`_.
The following code shows how to construct an instance of `kaldifeat.FbankOptions`_.
.. literalinclude:: ./code/fbank_options-1.txt
:caption: Usage of `kaldifeat.FbankOptions`_
:emphasize-lines: 6,8,22,37
:language: python
Note that we reuse the same option name with `compute-fbank-feats`_ from `Kaldi`_:
.. code-block:: bash
$ compute-fbank-feats --help
.. literalinclude:: ./code/compute-fbank-feats-help.txt
:caption: Output of ``compute-fbank-feats --help``
Please refer to the output of ``compute-fbank-feats --help`` for the meaning
of each field of `kaldifeat.FbankOptions`_.
One thing worth noting is that `kaldifeat.FbankOptions`_ has a field ``device``,
which is an instance of ``torch.device``. You can assign it either a string, e.g.,
``"cpu"`` or ``"cuda:0"``, or an instance of ``torch.device``, e.g., ``torch.device("cpu")`` or
``torch.device("cuda", 1)``.
.. hint::
You can use this field to control whether the feature computer
constructed from it performs computation on CPU or CUDA.
.. caution::
If you use a CUDA device, make sure that you have installed a CUDA version
of `PyTorch`_.
Example usage
-------------
The following code from
`<https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/tests/test_fbank_options.py>`_
demonstrate the usage of `kaldifeat.FbankOptions`_:
.. literalinclude:: ./code/test_fbank_options.py
:caption: Example usage of `kaldifeat.FbankOptions`_
:language: python

11
doc/source/usage/index.rst Normal file
View File

@ -0,0 +1,11 @@
Usage
=====
This section describes how to use feature computers in `kaldifeat`_.
.. toctree::
:maxdepth: 2
fbank_options
fbank
online_fbank

3
doc/source/usage/online_fbank.rst Normal file
View File

@ -0,0 +1,3 @@
kaldifeat.OnlineFbank
=====================

7
kaldifeat/csrc/CMakeLists.txt
View File

@ -31,9 +31,15 @@ function(kaldifeat_add_test source)
gtest_main
)
# NOTE: We set the working directory here so that
# it works also on windows. The reason is that
# the required DLLs are inside ${TORCH_DIR}/lib
# and they can be found by the exe if the current
# working directory is ${TORCH_DIR}\lib
add_test(NAME "Test.${name}"
COMMAND
$<TARGET_FILE:${name}>
WORKING_DIRECTORY ${TORCH_DIR}/lib
)
endfunction()
@ -47,4 +53,5 @@ if(kaldifeat_BUILD_TESTS)
foreach(source IN LISTS test_srcs)
kaldifeat_add_test(${source})
endforeach()
endif()

11
kaldifeat/python/csrc/CMakeLists.txt
View File

@ -10,6 +10,17 @@ pybind11_add_module(_kaldifeat
online-feature.cc
utils.cc
)
if(APPLE)
execute_process(
COMMAND "${PYTHON_EXECUTABLE}" -c "from distutils.sysconfig import get_python_lib; print(get_python_lib())"
OUTPUT_STRIP_TRAILING_WHITESPACE
OUTPUT_VARIABLE PYTHON_SITE_PACKAGE_DIR
)
message(STATUS "PYTHON_SITE_PACKAGE_DIR: ${PYTHON_SITE_PACKAGE_DIR}")
target_link_libraries(_kaldifeat PRIVATE "-Wl,-rpath,${PYTHON_SITE_PACKAGE_DIR}")
endif()
target_link_libraries(_kaldifeat PRIVATE kaldifeat_core)
if(UNIX AND NOT APPLE)
target_link_libraries(_kaldifeat PUBLIC ${TORCH_DIR}/lib/libtorch_python.so)

2
kaldifeat/python/csrc/feature-fbank.cc
View File

@ -53,7 +53,7 @@ static void PybindFbank(py::module &m) {
.def("dim", &PyClass::Dim)
.def_property_readonly("options", &PyClass::GetOptions)
.def("compute_features", &PyClass::ComputeFeatures, py::arg("wave"),
py::arg("vtln_warp"))
py::arg("vtln_warp"), py::call_guard<py::gil_scoped_release>())
.def(py::pickle(
[](const PyClass &self) -> py::dict {
return AsDict(self.GetOptions());

2
kaldifeat/python/csrc/feature-mfcc.cc
View File

@ -53,7 +53,7 @@ static void PybindMfcc(py::module &m) {
.def("dim", &PyClass::Dim)
.def_property_readonly("options", &PyClass::GetOptions)
.def("compute_features", &PyClass::ComputeFeatures, py::arg("wave"),
py::arg("vtln_warp"))
py::arg("vtln_warp"), py::call_guard<py::gil_scoped_release>())
.def(py::pickle(
[](const PyClass &self) -> py::dict {
return AsDict(self.GetOptions());

2
kaldifeat/python/csrc/feature-plp.cc
View File

@ -56,7 +56,7 @@ static void PybindPlp(py::module &m) {
.def("dim", &PyClass::Dim)
.def_property_readonly("options", &PyClass::GetOptions)
.def("compute_features", &PyClass::ComputeFeatures, py::arg("wave"),
py::arg("vtln_warp"))
py::arg("vtln_warp"), py::call_guard<py::gil_scoped_release>())
.def(py::pickle(
[](const PyClass &self) -> py::dict {
return AsDict(self.GetOptions());

2
kaldifeat/python/csrc/feature-spectrogram.cc
View File

@ -53,7 +53,7 @@ static void PybindSpectrogram(py::module &m) {
.def("dim", &PyClass::Dim)
.def_property_readonly("options", &PyClass::GetOptions)
.def("compute_features", &PyClass::ComputeFeatures, py::arg("wave"),
py::arg("vtln_warp"))
py::arg("vtln_warp"), py::call_guard<py::gil_scoped_release>())
.def(py::pickle(
[](const PyClass &self) -> py::dict {
return AsDict(self.GetOptions());

6
kaldifeat/python/csrc/online-feature.cc
View File

@ -22,9 +22,11 @@ void PybindOnlineFeatureTpl(py::module &m, const std::string &class_name,
.def_property_readonly("num_frames_ready", &PyClass::NumFramesReady)
.def("is_last_frame", &PyClass::IsLastFrame, py::arg("frame"))
.def("get_frame", &PyClass::GetFrame, py::arg("frame"))
.def("get_frames", &PyClass::GetFrames, py::arg("frames"))
.def("get_frames", &PyClass::GetFrames, py::arg("frames"),
py::call_guard<py::gil_scoped_release>())
.def("accept_waveform", &PyClass::AcceptWaveform,
py::arg("sampling_rate"), py::arg("waveform"))
py::arg("sampling_rate"), py::arg("waveform"),
py::call_guard<py::gil_scoped_release>())
.def("input_finished", &PyClass::InputFinished);
}

9
kaldifeat/python/kaldifeat/__init__.py
View File

@ -1,4 +1,13 @@
import torch
from .torch_version import kaldifeat_torch_version
if torch.__version__.split("+")[0] != kaldifeat_torch_version.split("+")[0]:
raise ImportError(
f"kaldifeat was built using PyTorch {kaldifeat_torch_version}\n"
f"But you are using PyTorch {torch.__version__} to run it"
)
from _kaldifeat import (
FbankOptions,
FrameExtractionOptions,

12
kaldifeat/python/kaldifeat/torch_version.py.in Normal file
View File

@ -0,0 +1,12 @@
# Auto generated by the toplevel CMakeLists.txt.
#
# DO NOT EDIT.
# The torch version used to build kaldifeat. We will check it against the
# torch version that is used to run kaldifeat. If they are not the same,
# `import kaldifeat` will throw.
#
# Some example values are:
# - 1.10.0+cu102
# - 1.5.0+cpu
kaldifeat_torch_version = "@TORCH_VERSION@"

13
kaldifeat/python/tests/Makefile Normal file
View File

@ -0,0 +1,13 @@
.PHONY: test
test:
python3 ./test_fbank.py
python3 ./test_fbank_options.py
python3 ./test_frame_extraction_options.py
python3 ./test_mel_bank_options.py
python3 ./test_mfcc.py
python3 ./test_mfcc_options.py
python3 ./test_plp.py
python3 ./test_plp_options.py
python3 ./test_spectrogram.py
python3 ./test_spectrogram_options.py

81
scripts/build_conda_cpu.sh Executable file
View File

@ -0,0 +1,81 @@
#!/usr/bin/env bash
#
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# The following environment variables are supposed to be set by users
#
# - KALDIFEAT_TORCH_VERSION
# The PyTorch version. Example:
#
# export KALDIFEAT_TORCH_VERSION=1.7.1
#
# Defaults to 1.7.1 if not set.
#
# - KALDIFEAT_CONDA_TOKEN
# If not set, auto upload to anaconda.org is disabled.
#
# Its value is from https://anaconda.org/kaldifeat/settings/access
# (You need to login as user kaldifeat to see its value)
#
# - KALDIFEAT_BUILD_TYPE
# If not set, defaults to Release.
set -e
export CONDA_BUILD=1
cur_dir=$(cd $(dirname $BASH_SOURCE) && pwd)
kaldifeat_dir=$(cd $cur_dir/.. && pwd)
cd $kaldifeat_dir
export KALDIFEAT_ROOT_DIR=$kaldifeat_dir
echo "KALDIFEAT_ROOT_DIR: $KALDIFEAT_ROOT_DIR"
KALDIFEAT_PYTHON_VERSION=$(python3 -c "import sys; print(sys.version[:3])")
if [ -z $KALDIFEAT_TORCH_VERSION ]; then
echo "env var KALDIFEAT_TORCH_VERSION is not set, defaults to 1.7.1"
KALDIFEAT_TORCH_VERSION=1.7.1
fi
if [ -z $KALDIFEAT_BUILD_TYPE ]; then
echo "env var KALDIFEAT_BUILD_TYPE is not set, defaults to Release"
KALDIFEAT_BUILD_TYPE=Release
fi
export KALDIFEAT_IS_FOR_CONDA=1
# Example value: 3.8
export KALDIFEAT_PYTHON_VERSION
# Example value: 1.7.1
export KALDIFEAT_TORCH_VERSION
export KALDIFEAT_BUILD_TYPE
if [ ! -z $KALDIFEAT_IS_GITHUB_ACTIONS ]; then
export KALDIFEAT_IS_GITHUB_ACTIONS
conda remove -q pytorch
conda clean -q -a
else
export KALDIFEAT_IS_GITHUB_ACTIONS=0
fi
if [ -z $KALDIFEAT_CONDA_TOKEN ]; then
echo "Auto upload to anaconda.org is disabled since KALDIFEAT_CONDA_TOKEN is not set"
conda build --no-test --no-anaconda-upload -c pytorch -c conda-forge ./scripts/conda-cpu/kaldifeat
else
conda build --no-test -c pytorch -c conda-forge --token $KALDIFEAT_CONDA_TOKEN ./scripts/conda-cpu/kaldifeat
fi

10
scripts/conda-cpu/cpuonly/meta.yaml Normal file
View File

@ -0,0 +1,10 @@
# this file is copied from
# https://github.com/pytorch/builder/tree/master/conda/cpuonly
package:
name: cpuonly
version: 1.0
build:
track_features:
- cpuonly
noarch: generic

33
scripts/conda-cpu/kaldifeat/build.sh Normal file
View File

@ -0,0 +1,33 @@
#!/usr/bin/env bash
#
# Copyright 2021 Xiaomi Corp. (author: Fangjun Kuang)
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
set -ex
CONDA_ENV_DIR=$CONDA_PREFIX
echo "KALDIFEAT_PYTHON_VERSION: $KALDIFEAT_PYTHON_VERSION"
echo "KALDIFEAT_TORCH_VERSION: $KALDIFEAT_TORCH_VERSION"
echo "KALDIFEAT_BUILD_TYPE: $KALDIFEAT_BUILD_TYPE"
echo "KALDIFEAT_BUILD_VERSION: $KALDIFEAT_BUILD_VERSION"
export KALDIFEAT_CMAKE_ARGS="-DCMAKE_BUILD_TYPE=${KALDIFEAT_BUILD_TYPE}"
export KALDIFEAT_MAKE_ARGS="-j1 VERBOSE=1"
export LIBRARY_PATH="/usr/local/miniconda/envs/kaldifeat/lib":$LIBRARY_PATH
export LD_LIBRARY_PATH="/usr/local/miniconda/envs/kaldifeat/lib":$LD_LIBRARY_PATH
export DYLD_LIBRARY_PATH="/usr/local/miniconda/envs/kaldifeat/lib":$DYLD_LIBRARY_PATH
python3 setup.py install --single-version-externally-managed --record=record.txt

44
scripts/conda-cpu/kaldifeat/meta.yaml Normal file
View File

@ -0,0 +1,44 @@
package:
name: kaldifeat
version: "1.16"
source:
path: "{{ environ.get('KALDIFEAT_ROOT_DIR') }}"
build:
number: 0
string: cpu_py{{ environ.get('KALDIFEAT_PYTHON_VERSION') }}_torch{{ environ.get('KALDIFEAT_TORCH_VERSION') }}
script_env:
- KALDIFEAT_IS_GITHUB_ACTIONS
- KALDIFEAT_TORCH_VERSION
- KALDIFEAT_PYTHON_VERSION
- KALDIFEAT_BUILD_TYPE
- KALDIFEAT_BUILD_VERSION
- KALDIFEAT_IS_FOR_CONDA
requirements:
build:
- {{ compiler('c') }} # [win]
- {{ compiler('cxx') }} # [win]
host:
- cmake=3.18
- python
- pytorch={{ environ.get('KALDIFEAT_TORCH_VERSION') }}
- gcc_linux-64=7 # [linux]
- cpuonly
- numpy
run:
- python
- pytorch={{ environ.get('KALDIFEAT_TORCH_VERSION') }}
- numpy
about:
home: https://github.com/csukuangfj/kaldifeat
license: Apache V2
license_file: LICENSE
summary: Kaldi-compatible feature extraction with PyTorch
description: |
Kaldi-compatible feature extraction with PyTorch,
supporting CUDA, batch processing, chunk processing, and autograd

2
scripts/conda/kaldifeat/build.sh
View File

@ -32,6 +32,6 @@ echo "gcc version: $($CC --version)"
echo "nvcc version: $(nvcc --version)"
export KALDIFEAT_CMAKE_ARGS="-DCMAKE_BUILD_TYPE=${KALDIFEAT_BUILD_TYPE}"
export KALDIFEAT_MAKE_ARGS="-j2"
export KALDIFEAT_MAKE_ARGS="-j3"
python3 setup.py install --single-version-externally-managed --record=record.txt

2
scripts/conda/kaldifeat/meta.yaml
View File

@ -1,6 +1,6 @@
package:
name: kaldifeat
version: "1.14"
version: "1.16"
source:
path: "{{ environ.get('KALDIFEAT_ROOT_DIR') }}"

145
scripts/github_actions/generate_build_matrix.py Executable file
View File

@ -0,0 +1,145 @@
#!/usr/bin/env python3
# Copyright 2022 Xiaomi Corp. (authors: Fangjun Kuang)
import argparse
import json
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument(
"--enable-cuda",
action="store_true",
default=False,
help="True to enable CUDA",
)
parser.add_argument(
"--for-windows",
action="store_true",
default=False,
help="True for windows",
)
parser.add_argument(
"--test-only-latest-torch",
action="store_true",
default=False,
help="""If True, we test only the latest PyTroch
to reduce CI running time.""",
)
return parser.parse_args()
def generate_build_matrix(enable_cuda, for_windows, test_only_latest_torch):
matrix = {
# 1.5.x is removed because there are compilation errors.
# See
# https://github.com/csukuangfj/k2/runs/2533830771?check_suite_focus=true
# and
# https://github.com/NVIDIA/apex/issues/805
# "1.5.0": {
# "python-version": ["3.6", "3.7", "3.8"],
# "cuda": ["10.1", "10.2"],
# },
# "1.5.1": {
# "python-version": ["3.6", "3.7", "3.8"],
# "cuda": ["10.1", "10.2"],
# },
"1.6.0": {
"python-version": ["3.6", "3.7", "3.8"],
"cuda": ["10.1", "10.2"]
if not for_windows
else ["10.1.243", "10.2.89"],
},
"1.7.0": {
"python-version": ["3.6", "3.7", "3.8"],
"cuda": ["10.1", "10.2", "11.0"]
if not for_windows
else ["10.1.243", "10.2.89", "11.0.3"],
},
"1.7.1": {
"python-version": ["3.6", "3.7", "3.8", "3.9"],
"cuda": ["10.1", "10.2", "11.0"]
if not for_windows
else ["10.1.243", "10.2.89", "11.0.3"],
},
"1.8.0": {
"python-version": ["3.6", "3.7", "3.8", "3.9"],
"cuda": ["10.1", "10.2", "11.1"]
if not for_windows
else ["10.1.243", "10.2.89", "11.1.1"],
},
"1.8.1": {
"python-version": ["3.6", "3.7", "3.8", "3.9"],
"cuda": ["10.1", "10.2", "11.1"]
if not for_windows
else ["10.1.243", "10.2.89", "11.1.1"],
},
"1.9.0": {
"python-version": ["3.6", "3.7", "3.8", "3.9"],
"cuda": ["10.2", "11.1"]
if not for_windows
else ["10.2.89", "11.1.1"],
},
"1.9.1": {
"python-version": ["3.6", "3.7", "3.8", "3.9"],
"cuda": ["10.2", "11.1"]
if not for_windows
else ["10.2.89", "11.1.1"],
},
"1.10.0": {
"python-version": ["3.6", "3.7", "3.8", "3.9"],
"cuda": ["10.2", "11.1", "11.3"]
if not for_windows
else ["10.2.89", "11.1.1", "11.3.1"],
},
"1.10.1": {
"python-version": ["3.6", "3.7", "3.8", "3.9"],
"cuda": ["10.2", "11.1", "11.3"]
if not for_windows
else ["10.2.89", "11.1.1", "11.3.1"],
},
"1.10.2": {
"python-version": ["3.6", "3.7", "3.8", "3.9"],
"cuda": ["10.2", "11.1", "11.3"]
if not for_windows
else ["10.2.89", "11.1.1", "11.3.1"],
},
"1.11.0": {
"python-version": ["3.7", "3.8", "3.9", "3.10"],
"cuda": ["10.2", "11.3", "11.5"]
if not for_windows
else ["11.3.1", "11.5.2"],
},
}
if test_only_latest_torch:
latest = "1.11.0"
matrix = {latest: matrix[latest]}
ans = []
for torch, python_cuda in matrix.items():
python_versions = python_cuda["python-version"]
cuda_versions = python_cuda["cuda"]
if enable_cuda:
for p in python_versions:
for c in cuda_versions:
ans.append({"torch": torch, "python-version": p, "cuda": c})
else:
for p in python_versions:
ans.append({"torch": torch, "python-version": p})
print(json.dumps({"include": ans}))
def main():
args = get_args()
generate_build_matrix(
enable_cuda=args.enable_cuda,
for_windows=args.for_windows,
test_only_latest_torch=args.test_only_latest_torch,
)
if __name__ == "__main__":
main()

7
scripts/github_actions/install_cuda.sh
View File

@ -36,6 +36,13 @@ case "$cuda" in
# url=https://developer.download.nvidia.com/compute/cuda/11.1.0/local_installers/cuda_11.1.0_455.23.05_linux.run
url=https://developer.download.nvidia.com/compute/cuda/11.1.1/local_installers/cuda_11.1.1_455.32.00_linux.run
;;
11.3)
# url=https://developer.download.nvidia.com/compute/cuda/11.3.0/local_installers/cuda_11.3.0_465.19.01_linux.run
url=https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.19.01_linux.run
;;
11.5)
url=https://developer.download.nvidia.com/compute/cuda/11.5.2/local_installers/cuda_11.5.2_495.29.05_linux.run
;;
*)
echo "Unknown cuda version: $cuda"
exit 1

35
scripts/github_actions/install_cudnn.sh
View File

@ -17,42 +17,43 @@
case $cuda in
10.0)
filename=cudnn-10.0-linux-x64-v7.6.5.32.tgz
url=http://www.mediafire.com/file/1037lb1vmj9qdtq/cudnn-10.0-linux-x64-v7.6.5.32.tgz/file
;;
10.1)
filename=cudnn-10.1-linux-x64-v8.0.2.39.tgz
url=http://www.mediafire.com/file/fnl2wg0h757qhd7/cudnn-10.1-linux-x64-v8.0.2.39.tgz/file
;;
10.2)
filename=cudnn-10.2-linux-x64-v8.0.2.39.tgz
url=http://www.mediafire.com/file/sc2nvbtyg0f7ien/cudnn-10.2-linux-x64-v8.0.2.39.tgz/file
;;
11.0)
filename=cudnn-11.0-linux-x64-v8.0.5.39.tgz
url=https://www.mediafire.com/file/abyhnls106ko9kp/cudnn-11.0-linux-x64-v8.0.5.39.tgz/file
;;
11.1)
filename=cudnn-11.1-linux-x64-v8.0.5.39.tgz
url=https://www.mediafire.com/file/qx55zd65773xonv/cudnn-11.1-linux-x64-v8.0.5.39.tgz/file
filename=cudnn-11.1-linux-x64-v8.0.4.30.tgz
;;
11.3)
filename=cudnn-11.3-linux-x64-v8.2.0.53.tgz
;;
11.5)
filename=cudnn-11.3-linux-x64-v8.2.0.53.tgz
;;
# 11.5)
# filename=cudnn-linux-x86_64-8.3.2.44_cuda11.5-archive.tar.xz
# ;;
*)
echo "Unsupported cuda version: $cuda"
exit 1
;;
esac
function retry() {
$* || (sleep 1 && $*) || (sleep 2 && $*) || (sleep 4 && $*) || (sleep 8 && $*)
}
command -v git-lfs >/dev/null 2>&1 || { echo >&2 "\nPlease install 'git-lfs' first."; exit 2; }
# It is forked from https://github.com/Juvenal-Yescas/mediafire-dl
# https://github.com/Juvenal-Yescas/mediafire-dl/pull/2 changes the filename and breaks the CI.
# We use a separate fork to keep the link fixed.
retry wget https://raw.githubusercontent.com/csukuangfj/mediafire-dl/master/mediafire_dl.py
git clone https://huggingface.co/csukuangfj/cudnn
cd cudnn
git lfs pull --include="$filename"
sed -i 's/quiet=False/quiet=True/' mediafire_dl.py
retry python3 mediafire_dl.py "$url"
sudo tar xf ./$filename -C /usr/local
rm -v ./$filename
sudo tar xf ./$filename --strip-components=1 -C /usr/local/cuda
# save disk space
git lfs prune && cd .. && rm -rf cudnn
sudo sed -i '59i#define CUDNN_MAJOR 8' /usr/local/cuda/include/cudnn.h

36
scripts/github_actions/install_torch.sh
View File

@ -78,7 +78,7 @@ case ${torch} in
;;
esac
;;
1.9.0)
1.9.*)
case ${cuda} in
10.2)
package="torch==${torch}"
@ -91,6 +91,40 @@ case ${torch} in
;;
esac
;;
1.10.*)
case ${cuda} in
10.2)
package="torch==${torch}"
# Leave it empty to use PyPI.
url=
;;
11.1)
package="torch==${torch}+cu111"
url=https://download.pytorch.org/whl/torch_stable.html
;;
11.3)
package="torch==${torch}+cu113"
url=https://download.pytorch.org/whl/torch_stable.html
;;
esac
;;
1.11.*)
case ${cuda} in
10.2)
package="torch==${torch}"
# Leave it empty to use PyPI.
url=
;;
11.3)
package="torch==${torch}+cu113"
url=https://download.pytorch.org/whl/torch_stable.html
;;
11.5)
package="torch==${torch}+cu115"
url=https://download.pytorch.org/whl/torch_stable.html
;;
esac
;;
*)
echo "Unsupported PyTorch version: ${torch}"
exit 1

35
scripts/github_actions/run-nightly-build.py Executable file
View File

@ -0,0 +1,35 @@
#!/usr/bin/env python3
# Copyright 2022 Xiaomi Corp. (authors: Fangjun Kuang)
import subprocess
from datetime import datetime, timedelta
def get_last_commit_date() -> datetime:
date = (
subprocess.check_output(
[
"git",
"log",
"-1",
"--format=%ad",
"--date=unix",
]
)
.decode("ascii")
.strip()
)
return datetime.utcfromtimestamp(int(date))
def main():
last_commit_date_utc = get_last_commit_date()
now_utc = datetime.utcnow()
if last_commit_date_utc + timedelta(days=1) > now_utc:
print("true")
else:
print("false")
if __name__ == "__main__":
main()

6
setup.py
View File

@ -61,5 +61,7 @@ with open("kaldifeat/python/kaldifeat/__init__.py", "r") as f:
with open("kaldifeat/python/kaldifeat/__init__.py", "w") as f:
for line in lines:
if "__version__" not in line:
f.write(line)
if "__version__" in line and "torch" not in line:
# skip __version__ = "x.x.x"
continue
f.write(line)