diff --git a/.github/scripts/unittest-windows/install.sh b/.github/scripts/unittest-windows/install.sh
index 73934b8fd0..8cc7c10df0 100644
--- a/.github/scripts/unittest-windows/install.sh
+++ b/.github/scripts/unittest-windows/install.sh
@@ -32,6 +32,8 @@ fi
 printf "Installing PyTorch\n"
 pip install --pre torch --index-url https://download.pytorch.org/whl/${UPLOAD_CHANNEL}/${wheel}
 
+python -c "import torch; print(torch.__version__)"
+
 torch_cuda=$(python -c "import torch; print(torch.cuda.is_available())")
 echo torch.cuda.is_available is $torch_cuda
 
@@ -42,6 +44,12 @@ if [ ! -z "${CUDA_VERSION:-}" ] ; then
     fi
 fi
 
+printf "Installing TorchCodec\n"
+# torchcodec nightly has no Windows+CUDA wheels, so we'll use CPU-only
+# torchcodec also under CUDA-enabled torch:
+pip install --pre torchcodec --index-url https://download.pytorch.org/whl/${UPLOAD_CHANNEL}/cpu
+python -c "import torchcodec; print(torchcodec.__version__)"
+
 # 2. Install torchaudio
 printf "* Installing fsspec\n"   # TODO: is this required for torchaudio??
 pip install --trusted-host pypi.org --trusted-host files.pythonhosted.org fsspec
diff --git a/.github/scripts/unittest-windows/run_test.sh b/.github/scripts/unittest-windows/run_test.sh
index 25d8e14196..292fe1d2b0 100644
--- a/.github/scripts/unittest-windows/run_test.sh
+++ b/.github/scripts/unittest-windows/run_test.sh
@@ -12,5 +12,9 @@ python -m torch.utils.collect_env
 env | grep TORCHAUDIO || true
 
 cd test
-pytest --continue-on-collection-errors --cov=torchaudio --junitxml=${RUNNER_TEST_RESULTS_DIR}/junit.xml -v --durations 20 torchaudio_unittest -k "not torchscript and not fairseq and not demucs and not librosa"
+if [ -z "${CUDA_VERSION:-}" ] ; then
+    pytest --continue-on-collection-errors --cov=torchaudio --junitxml=${RUNNER_TEST_RESULTS_DIR}/junit.xml -v --durations 20 torchaudio_unittest -k "not torchscript and not fairseq and not demucs and not librosa"
+else
+    pytest --continue-on-collection-errors --cov=torchaudio --junitxml=${RUNNER_TEST_RESULTS_DIR}/junit.xml -v --durations 20 torchaudio_unittest -k "not cpu and (cuda or gpu) and not torchscript and not fairseq and not demucs and not librosa"
+fi
 coverage html
diff --git a/.github/scripts/unittest-windows/setup_env.sh b/.github/scripts/unittest-windows/setup_env.sh
index 8203a1d978..efea89f548 100644
--- a/.github/scripts/unittest-windows/setup_env.sh
+++ b/.github/scripts/unittest-windows/setup_env.sh
@@ -36,4 +36,4 @@ fi
 conda activate "${env_dir}"
 
 # 3. Install minimal build tools
-conda install -y -c conda-forge cmake ninja
+conda install -y -c conda-forge cmake ninja "ffmpeg<8"
diff --git a/.github/workflows/unittest-windows-cpu.yml b/.github/workflows/unittest-windows-cpu.yml
index f6a786fcc1..3b872a4715 100644
--- a/.github/workflows/unittest-windows-cpu.yml
+++ b/.github/workflows/unittest-windows-cpu.yml
@@ -1,4 +1,4 @@
-name: Unittests on Windows CPU
+name: Unit-tests on Windows CPU
 
 on:
   pull_request:
@@ -10,7 +10,7 @@ on:
   workflow_dispatch:
 
 jobs:
-  unittests-windows-cpu:
+  build:
     uses: pytorch/test-infra/.github/workflows/windows_job.yml@main
     with:
       repository: pytorch/audio
@@ -21,7 +21,7 @@ jobs:
         git config --global --add safe.directory /__w/audio/audio
 
         # Set up Environment Variables
-        export PYTHON_VERSION="3.10"
+        export PYTHON_VERSION="3.11"
         export PIP_PROGRESS_BAR=off
         export CONDA_QUIET=1
         unset CUDA_VERSION
diff --git a/.github/workflows/unittest-windows-gpu.yml b/.github/workflows/unittest-windows-gpu.yml
index b00a512956..2b0e66c171 100644
--- a/.github/workflows/unittest-windows-gpu.yml
+++ b/.github/workflows/unittest-windows-gpu.yml
@@ -1,4 +1,4 @@
-name: Unittests on Windows GPU
+name: Unit-tests on Windows GPU
 
 on:
   pull_request:
@@ -10,7 +10,7 @@ on:
   workflow_dispatch:
 
 jobs:
-  unittests-windows-gpu:
+  build:
     uses: pytorch/test-infra/.github/workflows/windows_job.yml@main
     with:
       repository: pytorch/audio