Merge pull request csteinmetz1#45 from csteinmetz1/tests

Setting simple tests with Actions

.github/workflows/tests.yml

@@ -0,0 +1,27 @@
+name: pyloudnorm
+
+on: [push]
+
+jobs:
+  build:
+
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.7", "3.8", "3.9", "3.10"]
+
+    steps:
+      - uses: actions/checkout@v3
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v4
+        with:
+          python-version: ${{ matrix.python-version }}
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          sudo apt-get install libsndfile1-dev
+          pip install .[all]
+          pip install pytest
+      - name: Test with pytest
+        run: |
+          pytest

tests/test_auraloss.py

@@ -0,0 +1,89 @@
+import os
+import torch
+import auraloss
+
+
+def test_mrstft():
+    target = torch.rand(8, 2, 44100)
+    pred = torch.rand(8, 2, 44100)
+
+    loss = auraloss.freq.MultiResolutionSTFTLoss()
+    res = loss(pred, target)
+    assert res is not None
+
+
+def test_stft():
+    target = torch.rand(8, 2, 44100)
+    pred = torch.rand(8, 2, 44100)
+
+    loss = auraloss.freq.STFTLoss()
+    res = loss(pred, target)
+    assert res is not None
+
+
+def test_stft_weights_a():
+    target = torch.rand(8, 2, 44100)
+    pred = torch.rand(8, 2, 44100)
+    # test difference weights
+    loss = auraloss.freq.STFTLoss(
+        w_log_mag=1.0,
+        w_lin_mag=0.0,
+        w_sc=1.0,
+        reduction="mean",
+    )
+    res = loss(pred, target)
+    assert res is not None
+
+
+def test_stft_reduction():
+    target = torch.rand(8, 2, 44100)
+    pred = torch.rand(8, 2, 44100)
+    # test the reduction
+    loss = auraloss.freq.STFTLoss(
+        w_log_mag=1.0,
+        w_lin_mag=1.0,
+        w_sc=0.0,
+        reduction="none",
+    )
+    res = loss(pred, target)
+    print(res.shape)
+    assert len(res.shape) > 1
+
+
+def test_sum_and_difference():
+    target = torch.rand(8, 2, 44100)
+    pred = torch.rand(8, 2, 44100)
+    loss = auraloss.freq.SumAndDifferenceSTFTLoss()
+    res = loss(pred, target)
+    assert res is not None
+
+
+def test_melstft():
+    target = torch.rand(8, 2, 44100)
+    pred = torch.rand(8, 2, 44100)
+    # test MelSTFT
+    loss = auraloss.freq.MelSTFTLoss(44100)
+    res = loss(pred, target)
+    assert res is not None
+
+
+def test_melstft_reduction():
+    target = torch.rand(8, 2, 44100)
+    pred = torch.rand(8, 2, 44100)
+    # test MelSTFT with no reduction
+    loss = auraloss.freq.MelSTFTLoss(44100, reduction="none")
+    res = loss(pred, target)
+    assert len(res) > 1
+
+
+def test_multires_mel():
+    target = torch.rand(8, 2, 44100)
+    pred = torch.rand(8, 2, 44100)
+    sample_rate = 44100
+    loss = auraloss.freq.MultiResolutionSTFTLoss(
+        scale="mel",
+        n_bins=64,
+        sample_rate=sample_rate,
+    )
+    res = loss(pred, target)
+    assert res is not None