Merge master

.github/workflows/docker.yml

@@ -1,81 +1,124 @@
 name: docker
 on:
-  - workflow_dispatch
+  workflow_dispatch:
+    inputs:
+      version:
+        description: "Spleeter version to build image for"
+        required: true
+        default: "2.1.2"
 jobs:
-  build-test-push:
+  cuda-base:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        distribution: [3.6, 3.7, 3.8]
+      fail-fast: true
+    steps:
+      - uses: actions/checkout@v2
+      - name: Build CUDA base image
+        run: |
+          docker build \
+            --build-arg BASE=python:${{ matrix.distribution }} \
+            -t deezer/python-cuda-10-1:${{ matrix.distribution }} \
+            -f docker/cuda-10-1.dockerfile .
+      - name: Docker login
+        run: echo ${{ secrets.DOCKERHUB_PASSWORD }} | docker login -u ${{ secrets.DOCKERHUB_USERNAME }} --password-stdin
+      - name: Push deezer/python-cuda-10-1:${{ matrix.distribution }} image
+        run: docker push deezer/python-cuda-10-1:${{ matrix.distribution }}
+  pip-images:
+    needs: cuda-base
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        platform: [cpu, gpu]
+        distribution: [3.6, 3.7, 3.8]
+      fail-fast: true
+    steps:
+      - uses: actions/checkout@v2
+      - if: ${{ matrix.platform == 'cpu' }}
+        run: |
+          echo "base=python:${{ matrix.distribution }}" >> $GITHUB_ENV
+          echo "image=spleeter" >> $GITHUB_ENV
+      - if: ${{ matrix.platform == 'gpu' }}
+        run: |
+          echo "base=deezer/python-cuda-10-1:${{ matrix.distribution }}" >> $GITHUB_ENV
+          echo "image=spleeter-gpu" >> $GITHUB_ENV
+      - name: Build deezer/${{ env.image }}:${{ matrix.distribution }} image
+        run: |
+          docker build \
+            --build-arg BASE=${{ env.base }} \
+            --build-arg SPLEETER_VERSION=${{ github.event.inputs.version }} \
+            -t deezer/${{ env.image }}:${{ matrix.distribution }} \
+            -f docker/spleeter.dockerfile .
+      - name: Test deezer/${{ env.image }}:${{ matrix.distribution }} image
+        run: |
+          docker run \
+            -v $(pwd):/runtime \
+            deezer/${{ env.image }}:${{ matrix.distribution }} \
+            separate -o /tmp /runtime/audio_example.mp3
+      - name: Docker login
+        run: echo ${{ secrets.DOCKERHUB_PASSWORD }} | docker login -u ${{ secrets.DOCKERHUB_USERNAME }} --password-stdin
+      - name: Push deezer/${{ env.image }}:${{ matrix.distribution }} image
+        run: docker push deezer/${{ env.image }}:${{ matrix.distribution }}
+  conda-images:
+    needs: cuda-base
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        platform: [cpu, gpu]
+      fail-fast: true
+    steps:
+      - uses: actions/checkout@v2
+      - if: ${{ matrix.platform == 'cpu' }}
+        name: Build Conda base image
+        run: |
+          docker build -t conda:cpu -f docker/conda.dockerfile .
+          echo "image=spleeter" >> $GITHUB_ENV
+      - if: ${{ matrix.platform == 'gpu' }}
+        name: Build Conda base image
+        run: |
+          docker build --build-arg BASE=deezer/python-cuda-10-1:3.8 -t conda:gpu -f docker/conda.dockerfile .
+          echo "image=spleeter-gpu" >> $GITHUB_ENV
+      - name: Build deezer/${{ env.image }}:${{ env.tag }} image
+        run: |
+          docker build \
+            --build-arg BASE=conda:${{ matrix.platform }} \
+            --build-arg SPLEETER_VERSION=${{ github.event.inputs.version }} \
+            -t deezer/${{ env.image }}:conda \
+            -f docker/spleeter-conda.dockerfile .
+      - name: Docker login
+        run: echo ${{ secrets.DOCKERHUB_PASSWORD }} | docker login -u ${{ secrets.DOCKERHUB_USERNAME }} --password-stdin
+      - name: Push deezer/${{ env.image }}:conda image
+        run: docker push deezer/${{ env.image }}:conda
+  images-with-model:
+    needs: [pip-images, conda-images]
     runs-on: ubuntu-latest
     strategy:
       matrix:
         platform: [cpu, gpu]
         distribution: [3.6, 3.7, 3.8, conda]
-        model: [modelless, 2stems, 4stems, 5stems]
+        model: [2stems, 4stems, 5stems]
       fail-fast: true
     steps:
       - uses: actions/checkout@v2
-      # ----------------------------------------------------------------------
+      - if: ${{ matrix.platform == 'cpu' }}
-      # Note: base image building and env setup.
+        run: echo "image=spleeter" >> $GITHUB_ENV
-      - name: Setup Python distribution
-        run: |
-          echo "::set-env name=base::python:${{ matrix.distribution }}"
-          echo "::set-env name=tag::${{ matrix.distribution }}"
-          echo "::set-env name=file::spleeter"
-          echo "::set-env name=package::spleeter"
-      - if: ${{ matrix.distribution == 'conda' }}
-        name: Build Conda base image
-        run: |
-          docker build -t python:conda -f docker/conda.dockerfile .
-          echo "::set-env name=file::spleeter-conda"
       - if: ${{ matrix.platform == 'gpu' }}
-        name: Build CUDA base image
+        run: echo "image=spleeter-gpu" >> $GITHUB_ENV
+      - name: Build deezer/${{ env.image }}:${{ matrix.distribution }}-${{ matrix.model }} image
         run: |
           docker build \
-            --build-arg BASE=python:${{ matrix.distribution }} \
+            --build-arg BASE=deezer/${{ env.image }}:${{ matrix.distribution }} \
-            -t cuda:${{ matrix.distribution }} \
-            -f docker/cuda-10-0.dockerfile .
-          echo "::set-env name=base::cuda:${{ matrix.distribution }}"
-          echo "::set-env name=tag::${{ matrix.distribution }}-gpu"
-          echo "::set-env name=package::spleeter-gpu"
-      # ----------------------------------------------------------------------
-      # Note: image building.
-      - name: Build deezer/spleeter:${{ env.tag }} image
-        run: |
-          docker build \
-            --build-arg BASE=${{ env.base }} \
-            --build-arg SPLEETER_PACKAGE=${{ env.package }} \
-            -t deezer/spleeter:${{ env.tag }} \
-            -f docker/${{ env.file }}.dockerfile .
-          echo "::set-env name=modelargs::"
-      - if: ${{ matrix.model != 'modelless' }}
-        name: Build deezer/spleeter:${{ env.tag }}-${{ matrix.model }} image
-        run: |
-          docker build \
-            --build-arg BASE=deezer/spleeter:${{ env.tag }} \
             --build-arg MODEL=${{ matrix.model }} \
-            -t deezer/spleeter:${{ env.tag }}-${{ matrix.model }} \
+            -t deezer/${{ env.image }}:${{ matrix.distribution }}-${{ matrix.model }} \
             -f docker/spleeter-model.dockerfile .
-          echo "::set-env name=tag::${{ env.tag }}-${{ matrix.model }}"
+      - name: Test deezer/${{ env.image }}:${{ matrix.distribution }}-${{ matrix.model }} image
-          echo "::set-env name=modelarg::-p spleeter:${{ matrix.model }}"          
-      # ----------------------------------------------------------------------
-      # Note: image testing.
-      - name: Test deezer/spleeter:${{ env.tag }} image
         run: |
           docker run \
             -v $(pwd):/runtime \
-            deezer/spleeter:${{ env.tag }} \
+            deezer/${{ env.image }}:${{ matrix.distribution }} \
-            separate -i /runtime/audio_example.mp3  -o /tmp \${{ env.modelarg }}
+            separate -o /tmp -p spleeter:${{ matrix.model }} /runtime/audio_example.mp3
-      # ----------------------------------------------------------------------
-      # Note: image deploy.
       - name: Docker login
         run: echo ${{ secrets.DOCKERHUB_PASSWORD }} | docker login -u ${{ secrets.DOCKERHUB_USERNAME }} --password-stdin
-      - name: Push deezer/spleeter:${{ env.tag }} image
+      - name: Push deezer/${{ env.image }}:${{ matrix.distribution }}-${{ matrix.model }} image
-        run: docker push deezer/spleeter:${{ env.tag }}
+        run: docker push deezer/${{ env.image }}:${{ matrix.distribution }}-${{ matrix.model }}
-      - if: ${{ env.tag == 'spleeter:3.8' }}
-        name: Push deezer/spleeter:latest image
-        run: |
-          docker tag deezer/spleeter:3.8 deezer/spleeter:latest
-          docker push deezer/spleeter:latest
-      - if: ${{ env.tag == 'spleeter:3.8-gpu' }}
-        name: Push deezer/spleeter:gpu image
-        run: |
-          docker tag deezer/spleeter:3.8-gpu deezer/spleeter:gpu
-          docker push deezer/spleeter:gpu

.github/workflows/pypi.yml

@@ -1,8 +1,6 @@
 name: pypi
 on:
-  push:
+  - workflow_dispatch
-    branches:
-      - master
 env:
   PYPI_TOKEN: ${{ secrets.PYPI_TOKEN }}
 jobs:

README.md

@@ -1,6 +1,6 @@
 <img src="https://github.com/deezer/spleeter/raw/master/images/spleeter_logo.png" height="80" />
 
-[![Github actions](https://github.com/deezer/spleeter/workflows/pytest/badge.svg)](https://github.com/deezer/spleeter/actions) ![PyPI - Python Version](https://img.shields.io/pypi/pyversions/spleeter) [![PyPI version](https://badge.fury.io/py/spleeter.svg)](https://badge.fury.io/py/spleeter) [![Conda](https://img.shields.io/conda/vn/conda-forge/spleeter)](https://anaconda.org/conda-forge/spleeter) [![Docker Pulls](https://img.shields.io/docker/pulls/researchdeezer/spleeter)](https://hub.docker.com/r/researchdeezer/spleeter) [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/deezer/spleeter/blob/master/spleeter.ipynb) [![Gitter chat](https://badges.gitter.im/gitterHQ/gitter.png)](https://gitter.im/spleeter/community) [![status](https://joss.theoj.org/papers/259e5efe669945a343bad6eccb89018b/status.svg)](https://joss.theoj.org/papers/259e5efe669945a343bad6eccb89018b)
+[![Github actions](https://github.com/deezer/spleeter/workflows/pytest/badge.svg)](https://github.com/deezer/spleeter/actions) ![PyPI - Python Version](https://img.shields.io/pypi/pyversions/spleeter) [![PyPI version](https://badge.fury.io/py/spleeter.svg)](https://badge.fury.io/py/spleeter) [![Conda](https://img.shields.io/conda/vn/deezer-research/spleeter)](https://anaconda.org/deezer-research/spleeter) [![Docker Pulls](https://img.shields.io/docker/pulls/deezer/spleeter)](https://hub.docker.com/deezer/spleeter) [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/deezer/spleeter/blob/master/spleeter.ipynb) [![Gitter chat](https://badges.gitter.im/gitterHQ/gitter.png)](https://gitter.im/spleeter/community) [![status](https://joss.theoj.org/papers/259e5efe669945a343bad6eccb89018b/status.svg)](https://joss.theoj.org/papers/259e5efe669945a343bad6eccb89018b)
 
 > :warning: [Spleeter 2.1.0](https://pypi.org/project/spleeter/) release introduces some breaking changes, including new CLI option naming for input, and the drop
 > of dedicated GPU package. Please read [CHANGELOG](CHANGELOG.md) for more details.
@@ -45,16 +45,18 @@ Ready to dig into it ? In a few lines you can install **Spleeter** using [Conda]
 
 ```bash
 # install using conda
-conda install -c conda-forge spleeter
+conda config --add channels conda-forge # only needed if you don't already have this channel set
+conda install -c deezer-research spleeter 
 # download an example audio file (if you don't have wget, use another tool for downloading)
 wget https://github.com/deezer/spleeter/raw/master/audio_example.mp3
 # separate the example audio into two components
 spleeter separate -p spleeter:2stems -o output audio_example.mp3
 ```
+> :warning: for Mac Users, this will work but will install an old version of spleeter. To get the latest version, you need to install **Spleeter** using `pip`. Check the [wiki](https://github.com/deezer/spleeter/wiki/1.-Installation) for details.
 
 You should get two separated audio files (`vocals.wav` and `accompaniment.wav`) in the `output/audio_example` folder.
 
-For a detailed documentation, please check the [repository wiki](https://github.com/deezer/spleeter/wiki)
+For a detailed documentation, please check the [repository wiki](https://github.com/deezer/spleeter/wiki/1.-Installation)
 
 ## Development and Testing
 

conda/spleeter/meta.yaml

@@ -25,6 +25,7 @@ requirements:
     - python {{ python }}
     - tensorflow ==2.3.0  # [linux]
     - tensorflow ==2.3.0  # [win]
+    - numpy <1.20.0
     - pandas
     - ffmpeg-python
     - norbert

docker/spleeter-conda.dockerfile

@@ -2,12 +2,12 @@ ARG BASE=conda
 
 FROM ${BASE}
 
-ARG SPLEETER_PACKAGE=spleeter
 ARG SPLEETER_VERSION=1.5.3
 ENV MODEL_PATH /model
 
 RUN mkdir -p /model
+RUN conda config --add channels conda-forge
 RUN conda install -y -c conda-forge musdb
-RUN conda install -y -c conda-forge ${SPLEETER_PACKAGE}==${SPLEETER_VERSION}
+RUN conda install -y -c deezer-research spleeter 
 COPY docker/conda-entrypoint.sh spleeter-entrypoint.sh
 ENTRYPOINT ["/bin/bash", "spleeter-entrypoint.sh"]

docker/spleeter.dockerfile

@@ -2,13 +2,12 @@ ARG BASE=python:3.6
 
 FROM ${BASE}
 
-ARG SPLEETER_PACKAGE=spleeter
 ARG SPLEETER_VERSION=1.5.3
 ENV MODEL_PATH /model
 
 RUN mkdir -p /model
 RUN apt-get update && apt-get install -y ffmpeg libsndfile1
 RUN pip install musdb museval
-RUN pip install ${SPLEETER_PACKAGE}==${SPLEETER_VERSION}
+RUN pip install spleeter==${SPLEETER_VERSION}
 
 ENTRYPOINT ["spleeter"]

spleeter/dataset.py

@@ -241,7 +241,8 @@ class InstrumentDatasetBuilder(object):
     def filter_shape(self, sample):
         """ Filter badly shaped sample. """
         return check_tensor_shape(
-            sample[self._spectrogram_key], (self._parent._T, self._parent._F, 2)
+            sample[self._spectrogram_key],
+            (self._parent._T, self._parent._F, self._parent._n_channels),
         )
 
     def reshape_spectrogram(self, sample):
@@ -250,7 +251,8 @@ class InstrumentDatasetBuilder(object):
             sample,
             **{
                 self._spectrogram_key: set_tensor_shape(
-                    sample[self._spectrogram_key], (self._parent._T, self._parent._F, 2)
+                    sample[self._spectrogram_key],
+                    (self._parent._T, self._parent._F, self._parent._n_channels),
                 )
             },
         )
@@ -299,6 +301,7 @@ class DatasetBuilder(object):
         self._frame_length = audio_params["frame_length"]
         self._frame_step = audio_params["frame_step"]
         self._mix_name = audio_params["mix_name"]
+        self._n_channels = audio_params["n_channels"]
         self._instruments = [self._mix_name] + audio_params["instrument_list"]
         self._instrument_builders = None
         self._chunk_duration = chunk_duration
@@ -307,6 +310,21 @@ class DatasetBuilder(object):
         self._audio_path = audio_path
         self._random_seed = random_seed
 
+        self.check_parameters_compatibility()
+
+    def check_parameters_compatibility(self):
+        if self._frame_length / 2 + 1 < self._F:
+            raise ValueError(
+                "F is too large and must be set to at most frame_length/2+1. Decrease F or increase frame_length to fix."
+            )
+
+        if (
+            self._chunk_duration * self._sample_rate - self._frame_length
+        ) / self._frame_step < self._T:
+            raise ValueError(
+                "T is too large considering STFT parameters and chunk duratoin. Make sure spectrogram time dimension of chunks is larger than T (for instance reducing T or frame_step or increasing chunk duration)."
+            )
+
     def expand_path(self, sample):
         """ Expands audio paths for the given sample. """
         return dict(
@@ -368,7 +386,7 @@ class DatasetBuilder(object):
                 },
                 lambda x: tf.image.random_crop(
                     x,
-                    (self._T, len(self._instruments) * self._F, 2),
+                    (self._T, len(self._instruments) * self._F, self._n_channels),
                     seed=self._random_seed,
                 ),
             ),

spleeter/separator.py

@@ -307,7 +307,7 @@ class Separator(object):
         return prediction
 
     def separate(
-        self, waveform: np.ndarray, audio_descriptor: Optional[str] = None
+        self, waveform: np.ndarray, audio_descriptor: Optional[str] = ""
     ) -> None:
         """
         Performs separation on a waveform.

tests/test_train.py

@@ -52,17 +52,19 @@ TRAIN_CONFIG = {
 }
 
 
-def generate_fake_training_dataset(path, instrument_list=['vocals', 'other']):
+def generate_fake_training_dataset(path,
+                                   instrument_list=['vocals', 'other'],
+                                   n_channels=2,
+                                   n_songs = 2,
+                                   fs = 44100,
+                                   duration = 6,
+                                   ):
     """
         generates a fake training dataset in path:
         - generates audio files
         - generates a csv file describing the dataset
     """
     aa = AudioAdapter.default()
-    n_songs = 2
-    fs = 44100
-    duration = 6
-    n_channels = 2
     rng = np.random.RandomState(seed=0)
     dataset_df = pd.DataFrame(
         columns=['mix_path'] + [
@@ -83,26 +85,40 @@ def generate_fake_training_dataset(path, instrument_list=['vocals', 'other']):
 
 
 def test_train():
+
     with TemporaryDirectory() as path:
         # generate training dataset
-        generate_fake_training_dataset(path)
+        for n_channels in [1,2]:
-        # set training command aruments
+            TRAIN_CONFIG["n_channels"] = n_channels
+            generate_fake_training_dataset(path,
+                        n_channels=n_channels,
+                        fs=TRAIN_CONFIG["sample_rate"]
+                        )
+            # set training command arguments
             runner = CliRunner()
-        TRAIN_CONFIG['train_csv'] = join(path, 'train', 'train.csv')
+
-        TRAIN_CONFIG['validation_csv'] = join(path, 'train', 'train.csv')
+            model_dir = join(path, f'model_{n_channels}')
-        TRAIN_CONFIG['model_dir'] = join(path, 'model')
+            train_dir = join(path, f'train')
-        TRAIN_CONFIG['training_cache'] = join(path, 'cache', 'training')
+            cache_dir = join(path, f'cache_{n_channels}')
-        TRAIN_CONFIG['validation_cache'] = join(path, 'cache', 'validation')
+
+            TRAIN_CONFIG['train_csv'] = join(train_dir, 'train.csv')
+            TRAIN_CONFIG['validation_csv'] = join(train_dir, 'train.csv')
+            TRAIN_CONFIG['model_dir'] = model_dir
+            TRAIN_CONFIG['training_cache'] = join(cache_dir, 'training')
+            TRAIN_CONFIG['validation_cache'] = join(cache_dir, 'validation')
             with open('useless_config.json', 'w') as stream:
                 json.dump(TRAIN_CONFIG, stream)
+
             # execute training
             result = runner.invoke(spleeter, [
                 'train',
                 '-p', 'useless_config.json',
-            '-d', path
+                '-d', path,
+                "--verbose"
             ])
+
             # assert that model checkpoint was created.
-        assert os.path.exists(join(path, 'model', 'model.ckpt-10.index'))
+            assert os.path.exists(join(model_dir, 'model.ckpt-10.index'))
-        assert os.path.exists(join(path, 'model', 'checkpoint'))
+            assert os.path.exists(join(model_dir, 'checkpoint'))
-        assert os.path.exists(join(path, 'model', 'model.ckpt-0.meta'))
+            assert os.path.exists(join(model_dir, 'model.ckpt-0.meta'))
             assert result.exit_code == 0